Patent Application
20240020556
The present disclosure relates to the technical field of electronic devices, and in particular to methods and apparatus for information processing, a server, and a user device.
A black box algorithm is an algorithmic model established to solve complex problems that cannot be modeled using a clear mathematical formulation. The mathematical model of a black box algorithm is usually unknown, and the corresponding input data is input to an objective function of the black box algorithm to calculate and obtain output data corresponding to that input data. In the actual calculation process, the specific objective function of a black box algorithm is unknown. Due to the type of service and the high selectivity of black box algorithms, the black box algorithms can be applied to many fields such as e-commerce, finance, electricity, and so on.
In the existing technology, developers can implement black box optimization based on the Automated Machine Learning (AutoML) framework. The main process of black box optimization is to constantly generate, for parameters to be optimized, new samples by a parameter sampling algorithm, and then use an objective function of the black box to evaluate the effect of the use of the samples, to obtain the simulation results for the samples. After that, the parameter sampling algorithm is constantly used to generate new samples and perform simulation evaluation to generate a target sample that is close to the simulation result satisfying a predetermined condition as soon as possible. The target sample and the simulation result corresponding to the target sample are the optimization result of this black box optimization.
However, the existing black box optimization, after developed by developers, mainly performs the black box optimization process locally, resulting in low efficiency of black box optimization.
Embodiments of the present disclosure provide an information processing method. In some embodiments, the information processing method includes: detecting a parameter optimization request initiated by a target user; determining a parameter sampling algorithm matching the target user; invoking, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a test sample; determining a simulation result for the test sample based on a preset objective function; and outputting the simulation result for the test sample for the target user.
In some embodiments, the information processing method includes: receiving a parameter optimization request initiated by a target user for a parameter to be processed of a target resource; determining a parameter sampling algorithm matching a processing objective of the target resource; invoking, in response to a sample acquisition request initiated by the target user for the parameter to be processed, the parameter sampling algorithm to generate a test sample for the parameter to be processed; determining a simulation result for the test sample based on a preset objective function; and outputting for the target user the simulation result corresponding to the test sample for the parameter to be processed for the target user to process the target resource in accordance with the test sample.
In some embodiments, the information processing method includes: receiving a parameter optimization request for a browsing parameter corresponding to a browsing operation initiated by a target user; determining a parameter sampling algorithm matching a browsing objective of the target user; invoking, in response to a sample acquisition request initiated by the target user for the browsing parameter, the parameter sampling algorithm to generate a test sample for the browsing parameter; determining a simulation result for the test sample based on a preset objective function; and outputting for the target user the simulation result corresponding to the test sample for the browsing parameter for the target user to configure the browsing parameter in accordance with the test sample.
In some embodiments, the information processing method includes: detecting a parameter optimization request initiated by a target user for a parameter sampling algorithm; sending the parameter optimization request to a service end for the service end to determine a parameter sampling algorithm matching the target user; sending a sample acquisition request initiated by the target user to the service end for the service end to invoke, in response to the sample acquisition request, the parameter sampling algorithm to generate a test sample; determining a simulation result for the test sample based on a preset objective function; and outputting the simulation result for the test sample for the target user.
In some embodiments, the information processing method includes: determining, in response to a request to invoke an information processing interface, a processing resource corresponding to the information processing interface; and using the processing resource corresponding to the information processing interface to perform operations of: detecting a parameter optimization request initiated by a target user and determining a parameter sampling algorithm matching the target user; invoking, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a test sample; determining a simulation result for the test sample based on a preset objective function; and outputting the simulation result for the test sample for the target user.
Embodiments of the present disclosure provide an information processing apparatus. In some embodiments, the information processing apparatus includes: a request detection module configured to detect a parameter optimization request initiated by a target user and determine a parameter sampling algorithm matching the target user; a first response module configured to invoke, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a test sample; a first simulation module configured to determine a simulation result for the test sample based on a preset objective function; and a first output module configured to output the simulation result for the test sample for the target user.
In some embodiments, the information processing apparatus includes: a request initiation module configured to detect a parameter optimization request initiated by a target user for a parameter sampling algorithm; a first sending module configured to send the parameter optimization request provided by the target user to a service end for the service end to determine a parameter sampling algorithm matching the target user; a second sending module configured to send a sample acquisition request initiated by the target user to the service end for the service end to invoke, in response to the sample acquisition request, the parameter sampling algorithm to generate a test sample; a second simulation module configured to determine a simulation result for the test sample based on a preset objective function; and a second output module configured to output the simulation result for the test sample for the target user.
Embodiments of the present disclosure provide a server. The server includes: a storage component and a processing component, wherein the storage component is configured to store one or more computer instructions, the one or more computer instructions being invoked by the processing component; and the processing component is configured to: detect a parameter optimization request initiated by a target user and determine a parameter sampling algorithm matching the target user; invoke, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a test sample; determine a simulation result for the test sample based on a preset objective function; and output the simulation result for the test sample for the target user.
Embodiments of the present disclosure provide a user device. The user device includes: a storage component and a processing component, wherein the storage component is configured to store one or more computer instructions, the one or more computer instructions being invoked by the processing component; and the processing component is configured to: detect a parameter optimization request initiated by a target user for a parameter sampling algorithm; send the parameter optimization request provided by the target user to a service end for the service end to determine a parameter sampling algorithm matching the target user; send a sample acquisition request initiated by the target user to the service end for the service end to invoke, in response to the sample acquisition request, the parameter sampling algorithm to generate a test sample; determine a simulation result for the test sample based on a preset objective function; and output the simulation result for the test sample for the target user.
Embodiments of the present disclosure provide an electronic device. The electronic device includes: a memory configured to store one or more computer instructions; and one or more processors configured to run the one or more computer instructions stored in the memory, to execute operations of the above information processing methods.
Embodiments of the present disclosure provide a non-transitory computer-readable storage medium. The non-transitory computer-readable storage medium stores a set of instructions that are executable by one or more processors of a device to cause the device to perform operations of the above information processing methods.
In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or in the prior art, a brief description of the drawings required in the description of the embodiments or prior art will be given below, and it is clear that the drawings in the description below are some embodiments of the present disclosure, and that other drawings can be obtained from these drawings without any creative effort by a person of ordinary skill in the art.
In order to facilitate a better understanding of the objectives, technical solutions and advantages of the present disclosure, technical solutions in the embodiments of the present disclosure will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present disclosure. It is apparent that the described embodiments are only some of rather than all embodiments of the present disclosure. Based on the embodiments of the present disclosure, all other embodiments that can be derived by a person of ordinary skill in the art without any creative efforts should fall within the protection scope of the present disclosure.
The terms used in embodiments of the present disclosure are used solely for the purpose of describing particular embodiments and are not intended to limit the present disclosure. The singular forms “a,” “said,” and “the” used in the embodiments of the present disclosure and the appended claims are also intended to include plural forms, unless other meanings are clearly indicated in the context, and “multiple” generally includes at least two, but does not exclude the case of including at least one.
It should be understood that the term “and/or” used herein is only an association relationship describing associated objects, indicating that there may be three types of relationships. For example, A and/or B can mean three cases including that A exists alone, both A and B exist at the same time, and B exists alone. In addition, the character “/” here generally indicates that associated objects before and after it are in an “or” relationship.
Depending on the context, the words “in case of” and “if” as used herein may be interpreted as “at the time of” or “when” or “in response to determination” or “in response to identification”. Similarly, depending on the context, the phrases “if it is determined” or “if it is identified (stated condition or event)” may be interpreted as “when it is determined” or “in response to determining” or “when it is identified (the stated condition or event)” or “in response to identifying (the stated condition or event).”
It should also be noted that the terms “include”, “comprise”, or any other variation thereof are intended to cover non-exclusive inclusion, so that goods or systems including a series of elements include not only those elements, but also other elements not expressly listed, or elements that are inherent to such goods or systems. In the absence of more limitations, an element defined by “including a/an . . . ” does not exclude that the goods or systems including the element further include other identical elements.
Embodiments of the present disclosure provide methods and apparatus for information processing, a server, and a user device to solve the technical problem of low efficiency of black box optimization due to local execution of the black box optimization process in the conventional solutions.
In embodiments of the present disclosure, a parameter optimization request initiated by a target user can be received, and a parameter sampling algorithm matching the target user can be determined. Then, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm can be invoked to generate a test sample. The test sample may be used for determining a simulation result by way of calculation of an objective function, to output the simulation result for the test sample for the target user. Through interactions with the target user, a cloudification service for parameter optimization is implemented to achieve fast parameter optimization in the cloud, thereby improving the efficiency of parameter optimization. In addition, by acquiring the parameter optimization request and sample acquisition request from the user, the service demand of the user is acquired, and by providing effective response and feedback for the service demand of the user, the black box service is provided to the user effectively, so that the extended application of the black box service is realized, thus improving the utilization efficiency of the black box algorithm.
The technical solution of the embodiments of the present disclosure can be applied to the process of parameter optimization, and by providing external black box services, multi-user applications of black box optimization can be realized, thus extending the scope of application of black box optimization and improving the utilization efficiency of black box optimization.
In the conventional solutions, after developing black box algorithms based on the AutoML framework, developers can use the added black box optimization service. The main process of black box optimization is to use a black box algorithm to constantly generate new samples, and use the simulation function that simulates the actual calculation effect of users, i.e., the objective function, to simulate the use effect of the generated new samples to obtain the simulation results for the samples. By constantly generating new samples and constantly performing effect simulation of the newly generated samples, the sample with the highest use effect is selected as the global objective solution from many samples obtained. However, the existing black box algorithms, or black box optimization strategies are mostly used internally by developers, resulting in a narrow range of applications for black box optimization and poor utilization of black box algorithms.
In embodiments of the present disclosure, a parameter optimization request initiated by a target user can be detected, and a parameter sampling algorithm matching the target user can be determined. By providing external black box services, the target user can be enabled to use the black box algorithm for parameter optimization at any time. After invoking, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a test sample, it is possible to determine the simulation result for this test sample based on the preset objective function, and automatically generate a test sample for the target user and complete the simulation calculation for the test sample. The simulation result is generally the estimation result of the use effect of the test sample, to accurately evaluate the test sample, so that after outputted for the target user, the simulation result for the test sample can be viewed by the target user to determine whether the test sample is usable or not, thereby realizing the effect test of the test sample. Through interactions with the target user, a cloudification service for parameter optimization is implemented to achieve fast parameter optimization in the cloud, thereby improving the efficiency of parameter optimization. In addition, by effectively providing the black box service to the user, the extended application of the black box service is realized, thus improving the utilization efficiency of the black box algorithm.
Embodiments of the present disclosure will be described in detail below with the accompanying drawings.
In step 101, a parameter optimization request initiated by a target user is detected and a parameter sampling algorithm matching the target user is determined.
The information processing method according to some embodiments of the present disclosure can be applied to a service end. The service end is relative to the user end, and the service end and the user end correspond to a Client-Server (C/S) architecture. In practical applications, the service end can be, for example, a computer, a general server, a cloud server, a super personal computer, a laptop computer, a tablet computer, and so on. The user end can be configured on a user device. The user end can be, for example, a mobile phone, a tablet computer, a computer, a virtual reality device, an augmented reality device, a wearable device, or the like. Embodiments of the present disclosure do not limit the specific types of the service end and the user end.
Optionally, a plurality of parameter sampling algorithms can be configured in the service end. Different parameter sampling algorithms have different principles but produce the same results, and can all be used to generate test samples. The general parameter sampling algorithm can be used to generate a sample of a parameter to be optimized, and a preset objective function is used to simulate the use effect of the sample in order to predict the use effect of the sample and thus obtain the simulation result for the sample. By constantly generating samples and obtaining the simulation results for the samples, the target sample with the highest evaluation of the simulation results can be screened and the target sample can be outputted as the result of parameter optimization to complete the parameter optimization.
The target user can send the parameter optimization request to the service end through the user end. The service end can receive the parameter optimization request sent from the user end of the target user. The parameter optimization request can be generated by the target user based on the parameter to be optimized. Taking content recommendation in the field of e-commerce as an example, when a target user browses an e-commerce website, the website can push to the target user the content that is of interest to that target user. However, since there are various types of content and different types of content have different contributions to the click-through rate, the ratios of recommended content for various types can be used as parameters to be optimized, to perform parameter optimization.
The service end can be configured with multiple candidate sampling algorithms, and the target user can specify the parameter sampling algorithm among a plurality of black box algorithms of the parameter optimization request. At this point, the parameter optimization request may include algorithm identification information for the parameter sampling algorithm. The algorithm identification information may be, for example, the name of the algorithm.
In addition, the parameter optimization request may also include parameter information, such as the sample type, the value range, and the number, corresponding to samples of the parameter to be optimized. In some optional embodiments, the parameter sampling algorithm can generate a plurality of samples at the same time for the parameter to be optimized, to evaluate the plurality of samples at the same time and improve the efficiency of parameter optimization. In order to identify parameter optimization requests of different users, the parameter optimization request may also include a trial identifier that is set for this parameter optimization trial of the target user, to distinguish the optimization processes for different parameters. The target user can also log in to the server using user information such as the user name and password to obtain the access to the parameter optimization service, thus enabling effective management of the user.
Optionally, in practical applications, when receiving a parameter optimization request, the service end can store the parameter information, the user information, the trial identifier and other information in the parameter optimization request to facilitate accurate tracking of the process and result of the parameter optimization service, thereby improving the efficiency of information processing.
In step 102, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm is invoked to generate a test sample.
After determining the parameter sampling algorithm that matches the target user, the service end can feed back, to the target user, starting prompt information for starting parameter optimization. After sending the parameter optimization request, the target user can receive the starting prompt message corresponding to the parameter optimization that is fed back from the service end. Upon confirming the starting of the black box algorithm, the target user can send a sample acquisition request to the service end. The service end can receive the sample acquisition request sent from the target user, in response to this sample acquisition request, and invoke the parameter sampling algorithm to generate a test sample.
The test sample can be a parameter value that is set for the parameter to be optimized. Specifically, it can be a candidate solution generated by the parameter sampling algorithm for the parameter to be optimized, and when this candidate solution satisfies the optimization condition, it constitutes the final optimization result of the parameter to be optimized.
In step 103, a simulation result for the test sample is determined based on a preset objective function.
The objective function can be used to simulate the use effect or use result of the test sample to obtain the simulation result. The objective function can be determined by the optimization objective of the parameter to be optimized. For example, in the content recommendation scenario, the parameter to be optimized can be the parameter corresponding to the recommended content, and the test sample is the recommended content that is found according to the parameter to be optimized and shown to the browsing user. The higher the probability that the user clicks on the recommended content, the more effective the content recommendation is. Therefore, the prediction model for the click-through rate can be used as the objective function to predict the click-through rate of the recommended content by the browsing user, and the click-through rate obtained through the prediction is the simulation result for the test sample.
In step 104, the simulation result for the test sample is outputted for the target user.
Optionally, the operation of outputting the simulation result for the test sample for the target user may include: sending the simulation result for the test sample to the user end of the target user for the user end to display the output result for the test sample.
As a possible implementation, when the service end sends the simulation result for the test sample to the user end of the target user, it can be sent to the user end of the target user in the form of a message such as a web page, an instant messaging (IM) message, a short message service (SMS) message, or the like.
In embodiments of the present disclosure, a parameter optimization request initiated by a target user can be detected, and a parameter sampling algorithm matching the target user can be determined. By providing external black box services, the target user can be enabled to use the black box algorithm for parameter optimization at any time. After invoking, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a test sample, it can determine the simulation result for this test sample based on the preset objective function, and automatically generate a test sample for the target user and complete the simulation calculation for the test sample. Through interaction with the target user, a cloudification service for parameter optimization is implemented to achieve fast parameter optimization in the cloud, thereby improving the efficiency of parameter optimization. In addition, by acquiring the black box service request, the parameter optimization request, and the sample acquisition request from the user, the service demand of the user is acquired. By providing effective response and feedback for the service demand of the user, the black box service is provided to the user effectively, so that the extended application of the black box service can be realized, thus improving the utilization efficiency of the black box algorithm.
In some embodiments, the simulation result for the test sample is generally the evaluation result obtained by evaluating the use effect of the test sample based on the objective function. The simulation result can accurately evaluate the use effect of the test sample, so that after the simulation result for the test sample is outputted for the target user, the target user can view the simulation result for the test sample, to determine directly through the simulation result whether the test sample meets the requirements, thereby realizing the accurate prompt of the use effect of the test sample. For example, in the aforementioned content recommendation example, when the click-through rate calculation model is used as the objective function, the higher the click-through rate calculated and obtained is, the more effective the recommended content is. When the click-through rate reaches a threshold value set by the target user, the target user can confirm that the test sample meets the usage conditions. At this time, the test sample can be provided to the recommendation system so that the recommendation system can generate recommended content based on the test sample and recommend it to the user using the recommendation system.
In embodiments of the present disclosure, it can be open to various types of target users such as system developers, scholars, students, system designers, enterprise users, etc., to provide the parameter optimization service to different types of target users. When effectively providing the parameter optimization service to a target user, the target user can constantly initiate sample acquisition requests, and the service end can constantly respond to the sample acquisition requests to provide test samples to the target user, so that the target user can be informed of the simulation results for the test samples and select therefrom a target sample that meets their usage conditions. By providing the parameter optimization service to different target users, the extended application of the parameter optimization service can be achieved to enable more users to use the parameter optimization service, which improves the utilization efficiency of parameter optimization.
In embodiments of the present disclosure, the objective function can specifically be a black box algorithm, which cannot be expressed using a clear mathematical formula. The specific mathematical model of the objective function is usually unknown, and at this time, the parameter optimization request can include a black box optimization request.
In some embodiments, the operation of detecting a parameter optimization request initiated by the target user and determining a parameter sampling algorithm matching the target user may include: detecting a black box optimization request initiated by the target user, and determining the parameter sampling algorithm matching the target user.
The technical solution of embodiments of the present disclosure can be applied in a cloud server, which may include a public cloud and a private cloud. Among them, the public cloud may be a cloud provided on a third-party service for cloud users to use, which is typically used over a network to enable shared resource services and can serve an open public network. The private cloud is built for individual use by one customer and thus provides the most effective control over data, security, and quality of service. Therefore, the private cloud is more secure than the public cloud. For the sake of data security, the processing process can be different for public and private clouds when the user uses the information processing method configured at the cloud.
When the objective function is used to simulate test samples, more user information or data is involved. For the private cloud, when the objective function is used to simulate test samples, because the private cloud serves the user individually and has higher privacy, the simulation process for the test samples can be completed in the private cloud to reduce the processing pressure on the user end. For the public cloud, when the objective function is used to simulate test samples, because the public cloud serves many users and is not private, the simulation process for the test samples can be done at the user end in order to ensure the security of the data.
In one possible design, the public cloud and the private cloud can be implemented based on underlying services of K8S. K8S (Kubernetes, a container cluster management system) can be used to automatically deploy, expand and manage containerized applications. A cluster is a set of nodes. A cloud server, whether a public or private cloud, can be viewed as a service cluster. A service cluster can be formed by multiple nodes having the K8S platform installed on them to facilitate the management of the nodes in the cluster using the K8S. The information processing method in embodiments of the present disclosure can be applied to the K8S platform. Specifically, the information processing method can be configured on the custom resource controller (CRD) created in the K8S platform. That is, the CRD in the K8S can be used as a processing component to implement the information processing method.
The technical solution of embodiments of the present disclosure can be configured on the basis of the K8S underlying services, and the functions of each step can be modularized and then configured on the nodes. Interfaces for external services can be provided on the basis of the functional modules, and the data or information interaction with the user end can be realized through the external service interfaces to provide parameter optimization services to the users. For example, on the basis of the K8S, components such as a gateway service, a data storage component, a black box algorithm component, an early stop component, a web (World Wide Web) page, and a sample collection system, and a database, can be configured in the cloud server. The technical solution of embodiments of the present disclosure can be completed through the cooperation of these components.
In some embodiments, the gateway service can be a Rest (representational state transfer) API (application programming interface) gateway API server of the K8S. The data storage component can be a distributed storage component Etcd (storing cluster state information) of the K8S. The black box algorithm component can be a core algorithm component Suggestion Service provided for developers, which can include, for example, one or more of random search/grid search/cmaes (maximum likelihood estimation)/pma (partitioning around medoid). The early stop component can stop parameter simulation having unsatisfactory process as early as possible by early stop techniques, to reduce resource consumption and accelerate the test progress. The web page is mainly used for outwardly outputting test samples and simulation results of the test samples to users, allowing the users to view them in time. The sample collection system (metric system) can collect the black box test samples in the pod (scheduling unit). The database can be a database such as MySQL, which is used to store the test samples as well as data generated in the simulation and the simulation results.
When the cloud server is a public cloud, the simulation process of test samples can be completed at the user end to reduce the possibility of data leakage and improve data security.
In step 201, a parameter optimization request initiated by a target user is detected and a parameter sampling algorithm matching the target user is determined.
Some of the steps in embodiments of the present disclosure are the same as some of the steps illustrated in
Optionally, when the user end initiates a parameter optimization request to the service end, the request service information, which may include information such as algorithm information, parameter information, user information and/or verification information inputted at the user end, can be sent to the service end in the form of a Software Development Kit (SDK), and the information processing interface of the service end can receive the parameter optimization request encapsulated in an SDK form and obtain the parameter sampling algorithm corresponding to the request service information.
In step 202, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm is invoked to generate a test sample.
In step 203, the test sample to a user end of the target user for the user end to input the test sample to the preset objective function to calculate and obtain the simulation result corresponding to the test sample.
The test sample is sent to the user end of the target user and the simulation of the test sample is completed by the user end, and in this way, the risk of data leakage can be reduced and data security can be improved.
Optionally, before sending the test sample to the user end of the target user, the method may further include: receiving a sample test request sent from the user end of the target user, and sending the test sample to the user end of the target user in response to the sample test request.
In another possible design, test prompts can be provided to the target user. For example, test prompt information can be generated, which may include prompt information about the risk of performing the sample test in the public cloud, for example, in the form of text that reminds the target user that “there is a risk of data leakage, do you want to perform the simulation calculation of the test sample on the public cloud.” At this point, after receiving the risk prompt information, the user end can output the risk prompt information for the target user, so that the target user can view the risk prompt information and make corresponding feedback operations to the risk prompt information. In some embodiments, the target user can perform a confirmation operation for simulation of the test sample directly at the user end. When the service end detects the confirmation operation, the service end can send the test sample to the user end of the target user.
In step 204, the simulation result corresponding to the test sample sent is received by the user end.
In step 205, the simulation result for the test sample is outputted for the target user.
Optionally, the operation of outputting the simulation result for the test sample for the target user may include: controlling the target user to output the simulation result for the test sample. For example, when the service end receives the simulation result for the test sample, it can verify the simulation result, and after a successful verification, the service end can output the output control information for the simulation result to the user end. The user end can then output the simulation result in response to the output control information for the simulation result.
In embodiments of the present disclosure, the parameter optimization request initiated by the target user is detected, then the parameter sampling algorithm matching the target user can be determined, and in response to the sample acquisition request initiated by the target user, the parameter sampling algorithm can be invoked to generate a test sample. Accordingly, the test sample is sent to a user end of the target user for the user end to input the test sample to the preset objective function to calculate and obtain the simulation result corresponding to the test sample. By placing the testing work for the test sample at the user end, data transmission is reduced and the interaction efficiency is improved. After receiving the simulation result corresponding to the test sample sent from the user end, the simulation result for the test sample can be outputted for the target user.
When the cloud server is a private cloud, the simulation process for the test sample can be done at the user end to ensure the security of the data.
In step 301, a parameter optimization request initiated by a target user is detected and a parameter sampling algorithm matching the target user is determined.
Some of the steps in embodiments of the present disclosure are the same as some of the steps in the embodiments illustrated in
Optionally, when the user end initiates a parameter optimization request to the service end, the request service information, which may include information such as algorithm information, parameter information, user information and/or verification information inputted at the user end, can be sent to the service end in the form of a declarative API. Declarative means that the user end submits an API object that is defined for the request service information to declare the specific content and type of the desired request service information. The declarative API can contribute to the project orchestration capabilities of the K8S service, allowing a parameter optimization request to be interacted between the user end and the service end in a certain object format, to achieve accurate transmission of information.
When the service end receives the parameter optimization request, the service end can perform parsing in accordance with the type of information defined in the parameter optimization request as well as the amount of information, to obtain the relevant information in the parameter optimization request, thereby determining the parameter sampling algorithm that matches the target user.
It can be seen from the description in the foregoing embodiments that the black box algorithm component may actually include a plurality of classical algorithms, and the parameter sampling algorithm is one of them. In some embodiments, the parameter sampling algorithm may be directly specified by the target user, that is, at this time, the parameter optimization request may include the algorithm identification information for the parameter sampling algorithm. In some other embodiments, the parameter sampling algorithm may be determined by the service end in accordance with the parameter information provided by the target user, to select a black box algorithm that better matches the parameter information and improve the applicability of the parameter sampling algorithm to the current parameter optimization requirements.
In step 302, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm is invoked to generate a test sample.
In step 303, the test sample is inputted to the preset objective function to calculate and obtain the simulation result for the test sample.
Optionally, inputting the test sample to the preset objective function to calculate and obtain the simulation result for the test sample may include: determining a scheduling execution unit (e.g., a Pod) of the objective function, and inputting the test sample to the objective function in the scheduling execution unit for the scheduling execution unit to calculate and obtain the simulation result for the test sample. The scheduling execution unit is a computing module provided by the K8S service.
In step 304, the simulation result for the test sample is outputted for the target user.
In embodiments of the present disclosure, when the service end detects the parameter optimization request initiated by the target user and determines the parameter sampling algorithm matching the target user, the service end can invoke, in response to the sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a test sample, and then input the test sample input to the preset objective function to calculate and obtain the simulation result for the test sample, and output the simulation result for the test sample for the target user. Completing the simulation process for the test sample at the service end can reduce the processing pressure on the user end. In addition, because the computation efficiency of the service end is generally higher than that of the user end, the processing efficiency of test samples can be improved. By providing external black box services, the target user can use the black box algorithm for parameter optimization at any time, thus improving the efficiency of using the black box algorithm.
Optionally, when the service end is a cloud server, the cloud server can be obtained by integrating a plurality of nodes, and the parameter sampling algorithm can be run by one of the nodes. In some embodiments, after detecting a parameter optimization request initiated by the target user and determining a parameter sampling algorithm matching the target user, the method further includes: determining a target node for starting the parameter sampling algorithm. The operation of invoking, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a test sample includes: sending, in response to the sample acquisition request, the sample acquisition request to the target node starting the parameter sampling algorithm; and acquiring, through the target node, the test sample generated by the parameter sampling algorithm.
Optionally, the cloud server is based on a scalable server cluster, which may include a plurality of computing nodes. The target node for starting the parameter sampling algorithm can be determined by means of a node query Optionally, determining the target node for starting the parameter sampling algorithm may include: querying the target node for starting the black box algorithm from a plurality of nodes via a reverse proxy. As a possible implementation, the operation status of the black box algorithm in each node can be detected. Black box algorithms can be separately configured in different nodes, and one or more black box algorithms can be configured in any one node.
In a cloud server with K8S as the underlying service, the parameter sampling algorithm can be run in one of the nodes in the cloud server and the user end can actually acquire the test samples from the target node running the parameter sampling algorithm. Usually, the communication between the user end and the target node can be implemented through a router.
As a possible implementation, inputting the test sample to a preset objective function to calculate and obtain the simulation result for the test sample may include: establishing a simulation module for the objective function, and inputting the test sample to the simulation module to calculate and obtain the simulation result for the test sample.
The objective function depends on the support of the computing module during operation, so the simulation module is created for the objective function in order to enable the objective function to perform the simulation calculation of the test sample properly. In the K8S service, the simulation module may be a Pod.
The server may receive at the same time parameter optimization requests sent separately from a plurality of user ends, to serve the plurality of user ends at the same time. Alternatively, one user end may create different parameter optimization requests for different optimization requirements, and the server may also receive at the same time a plurality of parameter optimization requests sent from the one user end. Thus, in order to distinguish between different black box optimization processes, the target user can send a simulation trial identifier to the service end while initiating a parameter optimization request, to distinguish between different black box services.
Therefore, as a possible implementation, the method may further include: receiving a simulation trial identifier sent by the target user, and storing the simulation result for the test sample for the target user based on the simulation trial identifier.
Optionally, the operation of receiving the simulation trial identifier sent by the target user may specifically be: receiving the simulation trial identifier sent by the target user while detecting the parameter optimization request initiated by the target user. The simulation trial identifier is used to identify the simulation trial for the test sample in the parameter optimization request initiated by the target user. For example, when a target user A and a target user B initiate parameter optimization requests at the same time, the target user A may also send a simulation trial identifier “a” to the cloud server when initiating a parameter optimization request, and the target user B may also send a simulation trial identifier “b” to the cloud server when initiating a parameter optimization request. When the target user A initiates a sample acquisition request, a test sample generated for the target user A and a simulation result for the test sample can be identified by the simulation trial identifier “a.” When the target user B initiates a sample acquisition request, a test sample generated for the target user B and a simulation result for the test sample can be identified by the simulation trial identifier “b,” to distinguish them from the test sample generated for the target user A and the simulation result that are identified using the simulation trial identifier “a.”
In practical applications, when the service end provides a black box optimization service, the black box optimization process can be initiated based on the parameter sampling algorithm specified by the user.
In step 401, a parameter optimization request initiated by a target user is received.
Here, the parameter optimization request includes request service information for a parameter sampling algorithm.
Optionally, before receiving the parameter optimization request initiated by the target user, the method may further include: sending a plurality of black box algorithms to a user end of the target user for the user end to output the plurality of black box algorithms for the target user and select a parameter sampling algorithm from the plurality of black box algorithms. The request service information may include an algorithm name, algorithm identification, and other information that can be used to identify the black box algorithm.
In step 402, upon detecting the presence of a storage operation for the request service information in the data storage component, the parameter sampling algorithm for the target user to request service in the request service information is determined.
The request service information includes information such as algorithm information, parameter information, user information or verification information inputted at the user end. After the service end performs verification and other operations on the parameter optimization request, the request service information can be stored. In the K8S service, the request service information is usually stored using etcd (a distributed consistency protocol). The etcd is a distributed storage component developed in the kubenetes, which can store network configurations and object state information in a server cluster.
Optionally, the operation of the etcd to store the request service information may specifically be to detect a storage action of the etcd, and in the case of the presence of a storage action in the etcd, a storage message may be initiated to the processing component to enable the storage component to determine the presence of a storage operation for the request service information in the data storage component in a timely manner.
In step 403, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm is invoked to generate a test sample.
In step 404, a simulation result for the test sample is determined based on a preset objective function.
In step 405, the simulation result for the test sample is outputted for the target user.
Some of the steps in embodiments of the present disclosure are the same as some of the steps in the embodiments illustrated in
In embodiments of the present disclosure, the parameter optimization request initiated by the target user may include the request service information for the parameter sampling algorithm specified by the target user. The service end can detect the storage behavior of the data storage component, and can determine, upon detecting the presence of a storage operation for the request service information in the data storage component, the parameter sampling algorithm for the target user to request service in the request service information. The user specifies the black box service algorithm to achieve targeted services and enhance the user's controllability of the black box algorithm. After determining the parameter sampling algorithm, in response to the sample acquisition request initiated by the target user, the parameter sampling algorithm can be invoked to generate the test sample. Based on the preset objective function, the simulation result for the test sample is determined, to output the simulation result for the test sample for the target user. By providing external black box services, the target user can choose the parameter sampling algorithm for parameter optimization at any time, thus improving the efficiency of using the black box algorithm.
In practical applications, the request service information may also include verification information. The verification information may include identity verification information, permission verification information or the like for the target user.
In one possible design, after receiving the parameter optimization request sent by the target user, the method may also include: verifying the parameter optimization request of the target user based on the verification information in the request service information to obtain a verification result, and storing the request service information to the data storage component in response to the verification result being a successful verification.
Optionally, the verification information may include the identity verification information and the permission verification information for the target user. The operation of verifying the parameter optimization request of the target user based on the verification information in the request service information to obtain the verification result may include: verifying the user identity of the target user based on the identity verification information in the request service information to obtain an identity verification result; verifying the access permission of the target user based on the permission verification information in the request service information to obtain a permission verification result, and combining the identity verification result and the permission verification result to determine the verification result.
Further, optionally, the operation of combining the identity verification result and the permission verification result to determine the verification result may include: determining the verification result to be a successful verification in response to the identity verification result being a successful verification and the permission verification result being a successful verification; or determining the verification result to be a failed verification in response to the identity verification result being a failed verification or the permission verification result being a failed verification.
In some embodiments, in response to the verification result being a failed verification, verification failure prompt information is fed back to the user end of the target user for the user end to output the verification failure prompt information for the target user, to notify the user of the failure reason of the identity information or permission information, and to perform information modification or permission modification according to the failure reason, and re-perform the service application.
In yet some other embodiments, the operation of determining, upon detecting the presence of a storage operation for the request service information in a data storage component, the parameter sampling algorithm for the target user to request service in the request service information may include: acquiring, via a resource controller, a storage notification message initiated by the data storage component when storing the request service information, and determining, based on the storage notification message, the parameter sampling algorithm for the target user to request service in the request service information.
A resource controller may be configured in the service end, and the storage notification message initiated by the data storage component when storing the request service information is acquired via the resource controller. Through the storage notification message, the parameter sampling algorithm for the target user to request service in the request service information can be determined.
In yet some other embodiments, before invoking, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a test sample, the method may further include: receiving the sample acquisition request initiated by the target user for the parameter sampling algorithm.
In some embodiments, after determining a simulation result for the test sample based on a preset objective function, the method may further include: acquiring a judgment result as to whether the test sample satisfies a preset convergence condition; determining a target sample based on the test sample in response to the judgment result being that the test sample satisfies the convergence condition; or returning, in response to the judgment result being that the test sample does not satisfy the convergence condition, to the operation of invoking, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a test sample to continue execution.
Optionally, the operation of determining the target sample based on the test sample includes: acquiring historical test samples separately corresponding to N sample acquisition requests before the test sample, and selecting a target sample with the highest simulation result from the test sample and the N historical test samples based on simulation results separately corresponding to the test sample and the N historical test samples.
The simulation result may be specifically result indicator data. The higher the result indicator data, the better the simulation result, and the lower the result indicator data, the worse the simulation result.
In some embodiments, whether the test sample meets the preset convergence condition may be determined by whether the number of iterations corresponding to the test sample reaches a preset threshold number of iterations. If so, it is determined that the test sample meets the preset convergence condition. If not, it is determined that the test sample does not meet the preset convergence condition.
In some embodiments, whether the test sample meets the preset convergence condition may be determined by whether the simulation result for the test sample meets a preset result threshold. If so, it is determined that the test sample meets the preset convergence condition. If not, it is determined that the test sample does not meet the preset convergence condition. Whether the simulation result meets the preset result threshold may specifically include the simulation result being greater than the preset result threshold, or alternatively, the simulation result being less than the preset result threshold. Whether the simulation result is greater than the result threshold or less than the result threshold can be determined specifically according to the computational implications of the objective function.
To improve the efficiency of parameter processing, a plurality of test samples may be generated at a time in a batch processing manner.
In some embodiments, the operation of invoking, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a test sample may include: invoking, in response to the sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a plurality of test samples in a batch processing manner.
The operation of determining a simulation result for the test sample based on a preset objective function may include: determining simulation results separately corresponding to the plurality of test samples based on the preset objective function.
The operation of outputting the simulation result for the test sample for the target user may include: outputting the simulation results separately corresponding to the plurality of test samples for the target user.
Optionally, when determining simulation results separately corresponding to the plurality of test samples based on the preset objective function, if it is conducted on a public cloud, the plurality of test samples can be sent to the user end, and the sample simulation of the plurality of test samples can be completed at the user end. If it is conducted on a private cloud, the sample simulation of the plurality of test samples can be completed directly at the cloud.
Optionally, the operation of determining simulation results separately corresponding to the plurality of test samples based on the preset objective function specifically may include: inputting the plurality of test samples separately to the objective function to calculate and obtain the simulation results separately corresponding to the plurality of test samples.
In some embodiments, the step of acquiring simulation results can be performed on the cloud server or on the user end, which can be set according to the actual usage requirements. When any one of the service end or the user end obtains the simulation results, it can send the simulation results to the other end.
In one possible design, the operation of determining simulation results separately corresponding to the plurality of test samples based on the preset objective function may include: creating a plurality of simulation modules for the objective function; matching, from the plurality of simulation modules, target simulation modules separately for the plurality of test samples; and inputting any test sample to the target simulation module corresponding to that test sample to calculate and obtain a simulation result for the test sample to obtain the simulation results separately corresponding to the plurality of test samples.
In another possible design, the operation of determining simulation results separately corresponding to the plurality of test samples based on the preset objective function may include: sending the plurality of test samples to the user end of the target user for the user end to input the plurality of test samples separately to the preset objective function to calculate and obtain the simulation results separately corresponding to the plurality of test samples, and receiving the simulation results separately corresponding to the plurality of test samples that are sent by the user end.
In some embodiments, the method may further include: collecting the simulation results for the test samples by the SideCar collection approach in the K8S, and storing the simulation results for the test samples using the etcd.
The technical solution of embodiments of the present disclosure can be applied in various fields to solve the problem of parameter optimization. In the allocation process of electric power resources and water resources, the allocation result of electric power resources or water resources in each area can be used as a parameter to be processed to initiate a parameter optimization request, and the parameter to be processed can specifically be the amount of resources corresponding to each area, which, for example, may be the load capacity of the area in the electric power scenario.
In some embodiments, the operation of detecting a parameter optimization request initiated by the target user and determining a parameter sampling algorithm matching the target user may include: receiving a parameter optimization request initiated by a target user for a parameter to be processed of a target resource; determining a parameter sampling algorithm matching a processing objective of the target resource.
The operation of invoking, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a test sample includes: invoking, in response to a sample acquisition request initiated by the target user for the parameter to be processed, the parameter sampling algorithm to generate a test sample for the parameter to be processed. The operation of outputting the simulation result for the test sample for the target user includes: outputting for the target user a simulation result corresponding to the test sample for the parameter to be processed for the target user to process the target resource in accordance with the test sample.
Optionally, the target resources may include electric power resources, water resources, data resources, or the like. There are no excessive restrictions on the specific form and content of the resources in embodiments of the present disclosure.
The resource element specifically represented by the parameter to be processed may be determined according to the processing objective of the target resource. For example, the processing objective of the target resource can be the electric power load capacities set for different areas to minimize the total energy consumption of the grid. At this point, the electric power load capacities of different areas can be the parameters to be processed, and the processing objective can be the calculation function of the total energy consumption of the grid. The target parameter can be electric power load capacities of the various areas in the case of obtaining a minimum total energy consumption of the grid. According to the value of the parameter to be processed in the target parameter, the processing information for the target resource can be generated. That is, in accordance with the value of the parameter to be processed in the target parameter, prompt information or a setting instruction for the electric power load capacity of each area can be generated, and with the setting instruction, the capacity setting can be performed according to the electric power load capacity of each area. The prompt information can be displayed to the user for the user to perform capacity setting for each area according to the electric power load capacity of the area indicated in the prompt information.
In the field of e-commerce, the parameter optimization problem is also involved. Take the common product recommendation as an example, due to different browsing characteristics of users, such as the consumption habits, fields of interest, and historical browsing behaviors, the content or products recommended for users are also different. In practical applications, in order to improve the user's click-through rate, the user's browsing characteristics such as consumption habits and fields of interest can be parameterized to generate different browsing parameters, and through the setting of the plurality of browsing parameters, accurate analysis of the characteristics of the user's click-through objective can be achieved, to find the target products with higher user attention. The solution for performing parameter sampling of a plurality of browsing parameters and conducting parameter trials to determine the click-through rate of the user can be applied to the technical solution of embodiments of the present disclosure to improve the efficiency of trials.
Therefore, in some embodiments, the operation of detecting a parameter optimization request initiated by the target user and determining a parameter sampling algorithm matching the target user includes: receiving a parameter optimization request for a browsing parameter generated for a browsing operation initiated by the target user; and determining a parameter sampling algorithm matching a browsing objective of the target user.
Optionally, the operation of invoking, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a test sample includes: invoking, in response to a sample acquisition request initiated by the target user for the browsing parameter, the parameter sampling algorithm to generate a test sample for the browsing parameter.
The operation of outputting the simulation result for the test sample for the target user includes: outputting for the target user the simulation result corresponding to the test sample for the browsing parameter for the target user to configure the browsing parameter in accordance with the test sample.
In step 501, a parameter optimization request initiated by a target user for a parameter sampling algorithm is detected.
The technical solution of embodiments of the present disclosure can be applied to the user end. The user end can be, for example, an electronic device such as a mobile phone terminal, a computer, a laptop, a tablet computer, a virtual reality/augmented reality device, and an Internet of Things (IoT) terminal. There are no excessive restrictions on the specific type of the electronic device in embodiments of the present disclosure.
The specific contents performed by the user end as well as the technical effect have been described in detail in the aforementioned embodiments, and will not be repeated here for brevity.
In step 502, the parameter optimization request is sent to a service end for the service end to determine a parameter sampling algorithm matching the target user.
In step 503, a sample acquisition request initiated by the target user is sent to the service end for the service end to invoke, in response to the sample acquisition request, the parameter sampling algorithm to generate a test sample.
In step 504, a simulation result for the test sample is determined based on a preset objective function.
In step 505, the simulation result for the test sample is outputted for the target user.
Optionally, the objective function may be a black box algorithm, and in this case, the parameter optimization request is a black box optimization request.
In embodiments of the present disclosure, the user end can detect the parameter optimization request initiated by the target user for the parameter sampling algorithm, and then send the parameter optimization request to the service end. Upon receiving the parameter optimization request sent from the user end, the service end can determine the parameter sampling algorithm corresponding to the target user. After that, the user end can also send the sample acquisition request initiated by the target user to the service end. Upon receiving the sample acquisition request, the service end can invoke the parameter sampling algorithm to generate the test sample, and determine the simulation result for the test sample based on the preset objective function, and output the simulation result for the test sample for the target user. The user end provides the work of generation of sample parameters and simulation of sample parameters for the target user, and through the interaction with the target user, the parameter optimization service is realized. Thus, the target user can choose the parameter sampling algorithm for parameter optimization at any time, which improves the efficiency of using the black box algorithm.
Optionally, the operation of determining the simulation result for the test sample based on the preset objective function may include: receiving the test sample sent by the service end, and inputting the test sample to the objective function to calculate and obtain the simulation result for the test sample.
Optionally, the operation of determining the simulation result for the test sample based on the preset objective function may include: receiving the simulation result for the test sample sent by the service end, in which the simulation result is calculated and obtained by inputting the test sample to the simulation function.
In some embodiments, the operation of determining the simulation result for the test sample based on the preset objective function may include: receiving the test sample sent by the service end, and inputting the test sample to the objective function to calculate and obtain a calculation result corresponding to the test sample.
In one possible design, the operation of receiving the test samples sent by the service end may include: receiving a plurality of test samples sent by the service end, in which the plurality of test samples are generated by the service end by invoking the parameter sampling algorithm in a batch processing manner.
The operation of determining the simulation result for the test sample based on the preset objective function may include: inputting the plurality of test samples separately to the objective function to calculate and obtain calculation results separately corresponding to the plurality of test samples.
In some other embodiments, the operation of determining the simulation result for the test sample based on the preset objective function includes: receiving the simulation result corresponding to the test sample sent by the service end, in which the simulation result is calculated and obtained by inputting the test sample to the preset objective function by the service end.
In some embodiments, after determining a simulation result for the test sample based on a preset objective function, the method may further include: acquiring a judgment result as to whether the test sample satisfies a preset convergence condition; determining a target sample based on the test sample in response to the judgment result being that the test sample satisfies the convergence condition; or returning, in response to the judgment result being that the test sample does not satisfy the convergence condition, to the operation of sending a sample acquisition request initiated by the target user to the service end for the service end to invoke, in response to the sample acquisition request, the parameter sampling algorithm to generate a test sample to continue execution.
Optionally, the operation of determining the target sample based on the test sample may include: acquiring historical test samples separately corresponding to N sample acquisition requests before the test sample, and selecting a target sample with the highest simulation result from the test sample and the N historical test samples based on simulation results separately corresponding to the test sample and the N historical test samples.
The simulation result can be result indicator data, the higher the result indicator data, the better the simulation result, and the smaller the result indicator data, the worse the simulation result.
It should be noted that some of the steps in embodiments of the present disclosure, when executed at the user end or executed at the service end, perform the same specific steps, and the specific implementation of each step and the technical effect have been described in detail in the preceding embodiments and will not be repeated here for brevity.
In step 601, in response to a request to invoke an information processing interface, a processing resource corresponding to the information processing interface is determined.
The processing resource corresponding to the information processing interface is used to perform the steps 602-605.
In step 602, a parameter optimization request initiated by a target user is detected and a parameter sampling algorithm matching the target user is determined.
In step 603, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm is invoked to generate a test sample.
In step 604, a simulation result for the test sample is determined based on a preset objective function.
In step 605: the simulation result for the test sample is outputted for the target user.
The specific steps performed by the processing resources corresponding to the information processing interface in embodiments of the present disclosure are the same as the processing steps performed by the information processing methods illustrated in
In practical applications, the user end of the target user can be, for example, an electronic device such as a mobile phone terminal, a computer, a laptop, a tablet computer, a virtual reality/augmented reality device, and an Internet of Things (IoT) terminal. There are no excessive restrictions on the specific type of the electronic device in embodiments of the present disclosure. The user end can interact with the user, and the user end can communicate with a server capable of optimizing the parameters.
For ease of understanding, an application example of embodiments of the present disclosure will be introduced in detail, by taking an example that a hyper-parameter including the network depth, the number of iterations, and the number of neurons per layer of the machine learning model is the parameter that needs to be optimized, and that the cloud service end provides the information processing method and interacts with the user end.
Referring to
After that, the mobile phone terminal M2 can, in step 704, detect the sample acquisition request initiated by the target user and send the sample acquisition request 705 to the cloud server M2.
After that, the cloud server M2 can, in step 706, invoke, in response to the sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a test sample. The cloud server M2 can also, in step 707, determine the simulation result for the test sample based on the preset objective function, and, in step 708, output the simulation result for the test sample for the target user.
Optionally, operations of the cloud server M2 outputting the simulation result for the test sample for the target user may include: in step 7081, generating a prompt page for the simulation result based on the test sample, and sending the prompt page 7082 to the mobile phone terminal M1. After the mobile phone terminal M1 receives the prompt page, it can display the prompt page 7083 on a display screen.
In addition, the simulation result for the test sample can also be outputted in the form of data strings, SMS information or instant messaging (IM) messages, or the like. There are no excessive restrictions on the specific output manner of the simulation result in embodiments of the present disclosure.
The technical solution of embodiments of the present disclosure can be applied in a variety of fields such as artificial intelligence interaction, data retrieval, content recommendation, click-through rate prediction, intelligent factory, industrial control, and so on, and has stronger applicability especially in the field of content recommendation, such as content recommendation in the field of e-commerce, live video broadcasting, social networking, and online education, and in the field of resource allocation, such as financial product allocation, electric power resources, water resources, supply chain allocation.
For ease of understanding, the following problem cases in the actual field scenarios are used as examples to introduce embodiments of the present disclosure in detail.
First, for the field of e-commerce, the application scenarios such as product recommendation, content recommendation and calculation of advertisement click-through rates in the feature search and live broadcast scenarios are most common in the field of e-commerce, and the following embodiments provide an example deployment with a content recommendation scenario as an example. The general recommendation process in the recommendation scenario can be as follows: performing parameterization setting of the elements of the selected scenario, obtaining parameters that influence the scenario, and using the parameters to identify different features of the scenario. The service end can show a plurality of black box algorithms for the target user and describe in detail the functions and the effect of each black box algorithm. The target user can select a parameter sampling algorithm from the plurality of black box algorithms, and initiate a parameter optimization request to the service end through the user end. When the service end obtains the parameter optimization request initiated by the target user, it can determine the parameter sampling algorithm selected by the target user.
During the optimization process of a parameter, several trials of the parameter are required to obtain the target parameter. In the process of any one parameter trial, the target user can initiate a sample acquisition request corresponding to a parameter matching the recommendation scenario through the user end. At this time, the service end can respond to the sample acquisition request initiated by the target user and invoke the parameter sampling algorithm to generate a test sample. After that, a simulation result for the test sample is determined based on the preset objective function. After the service end determines the simulation result, it can send the simulation result to the user end, and the user end shows the simulation result for the test sample for the target user. By viewing the simulation result, the target user can judge whether the test sample can be used or not. If the test sample is judged by the target user to meet the usage conditions, the target user can use the test sample to set the recommendation elements and recommend content to the user according to the recommendation elements.
As an example, in the click word recommendation scenario, when a user clicks the query box in an application (APP), the system will recommend some query terms for that user. The purpose of recommending query terms for the user is to explore their potential purchase needs, increase the user's stickiness and increase the total number of goods transactions. The search system uses the following architecture, which combines a deep learning Encode-Decode network, i.e., a target network, to predict the recommendation of query terms. Assuming that the selection of target parameters is made for the number of query terms, in the conventional solutions, the parameter value of the parameter to be optimized including the number of query terms is set manually according to human experience. Using the data processing methods of embodiments of the present disclosure, parameter trials can be performed automatically on the number of query terms according to the above parameter optimization process. The test sample obtained by the service end is the number of query terms, and the simulation result for the test sample is the click-through rate or transaction rate obtained after the setting according to that number of query terms.
Second, for the social field, content recommendation for social users and material recommendation for students are also common in the social field. Recommendations in the social field are usually that social users browse social applications, and the applications display social content of interest to the users in the display interface. Generally, recommendations in the social field are usually that options such as historical browsing behavior, areas of interest, and user information of users form feature parameters, and combinations of different options can form parameters to be processed. When test samples corresponding to the parameters to be processed are determined, the recommendation of social content can be made according to the test samples. In order to find the content of interest to social users, the number and the type of parameters can be optimized to obtain accurate social user content for testing.
The technical solution of embodiments of the present disclosure can be configured in the cloud server, and the parameter optimization request can be initiated by the operation and maintenance personnel, who can use the user information, social type, and so on, as the parameters to be processed, and constantly generate test samples using the parameter sampling algorithm. Simulation results for test parameters are obtained by simulating the use effect of the test samples, to select target samples from a plurality of test samples. Further, the feature parameters used for social content recommendation are set according to the target samples, to achieve accurate recommendations.
Third, for the financial field, simulation of stock indices is a very important problem, and models based on linear regression, support vector machines (SVM) and Long Short-Term Memory (LSTM), etc., are common for the simulation problem of stock indices. Before the model is used, a suitable model needs to be established. Here, the model training process may involve various hyper-parameters, such as time_step (time step), feature_dim (feature dimension), hidden feature, and so on in LSTM, and may also involve contextual features, such as the macro-factors, the micro-factors and emergencies of the market, which will affect the selection of parameters. Using the technology of embodiments of the present disclosure, test samples can be generated for hyper-parameters that need to be set in the financial field, and corresponding simulation results can be obtained by performing simulations on the test samples.
The simulation results are outputted to the target user for viewing, so that when selecting a target sample with good simulation results from many test samples, the target user uses the target sample to construct a machine learning model corresponding to the index simulation problem, and perform model training to obtain the model parameters. Then, the machine learning model obtained through training is used to perform simulation and calculation of data such as the Root Mean Squared Error (RMSE) difference values of the actual stock index for the index simulation problem.
Fourth, for the field of resource allocation, the allocation of electric power resources is taken as an example. The allocation of electric power resources usually involves many regions, and each region can be represented using the corresponding parameters. The parameters can separately allocate a certain percentage of resources, and the allocation of resources can affect the regional economy, population, environment and other information.
The technical solution of embodiments of the present disclosure can be applied to the problem of dynamic pricing in the electric power market and the problem of economic load allocation of electric power. The following mainly provides a detailed description of the specific application fields of electric power systems.
In the dynamic pricing problem of the electric power market, the impact of the user type and the amount of electricity consumption on the electric power market is critical, and a parameter optimization request can be initiated taking parameters such as the user type and the amount of electricity consumption as parameters to be processed. In response to the parameter optimization request, a parameter sampling algorithm is determined and then used to generate test samples for the parameters to be processed. The setting of candidate parameters for the benefit/cost in the electric power system can be used as the final optimization objective to determine the objective function corresponding to this optimization objective. Using the objective function, the use result of the test sample can be simulated to obtain a simulation result, which represents the value of the benefit/cost of the system. After that, a target sample can be selected from many test samples by constantly generating test samples and simulating the test samples.
In the problem of economic load distribution of electric power, the electric power supplier can provide electric power resources to a plurality of areas at the same time, the electric power load capacity of each area can be used as a candidate parameter, and the total energy consumption of the grid can be used as the output of the objective function. Using the technical solution of embodiments of the present disclosure, the target user can initiate a parameter optimization request taking the respective electric power load capacity of each area as the parameter to be processed. After the parameter sampling algorithm matching the parameter optimization request of the target user is determined, the sample acquisition request initiated by the target user can be received to invoke the parameter sampling algorithm to constantly review test samples corresponding to the test value of the electric load capacity of each area. In the field of load distribution of electric power, the objective function may be a nonlinear constraint relationship between the electric load capacity and the total energy consumption of the grid. The simulation result can be obtained by simulating the test sample using the objective function. After that, simulations are constantly performed on the test samples of the electric power load capacity of each area, and the highest simulation result is selected from them. The test sample corresponding to the highest simulation result can be used to generate the load capacity allocation strategy.
For ease of understanding, the technical solution of embodiments of the present disclosure is described in detail by taking the field of e-commerce as an example.
In step 801, a parameter optimization request initiated by a target user for a parameter to be processed of a target resource is received.
In step 802, a parameter sampling algorithm matching a processing objective of the target resource is determined.
In step 803, in response to a sample acquisition request initiated by the target user for the parameter to be processed, the parameter sampling algorithm is invoked, to generate a test sample for the parameter to be processed.
In step 804, a simulation result for the test sample is determined based on a preset objective function.
In step 805, a simulation result corresponding to the test sample for the parameter to be processed is outputted to the target user, for the target user to process the target resource in accordance with the test sample.
In the allocation process of electric power resources and water resources, the allocation result of electric power resources or water resources in each area can be used as a parameter to be processed to initiate a parameter optimization request, and the parameter to be processed can specifically be the amount of resources corresponding to that each area, which, for example, may be the load capacity of the area in the electric power scenario.
Optionally, the target resources may include electric power resources, water resources, data resources, or the like. There are no excessive restrictions on the specific form and content of the resources in embodiments of the present disclosure.
The resource element specifically represented by the parameter to be processed may be determined according to the processing objective of the target resource. For example, the processing objective of the target resource can be the electric power load capacities set for different areas to minimize the total energy consumption of the grid. At this point, the electric power load capacities of different areas can be the parameters to be processed, and the processing objective can be the calculation function of the total energy consumption of the grid. The target parameters can be electric power load capacities of the various areas in the case of obtaining the minimum total energy consumption of the grid. According to the value of the parameter to be processed in the target parameters, the processing information for the target resource can be generated. That is, in accordance with the value of the parameter to be processed in the target parameters, prompt information or a setting instruction for the electric power load capacity of each area can be generated, and with the setting instruction, the capacity setting can be performed according to the electric power load capacity of each area. The prompt information can be displayed to the user for the user to perform capacity setting for each area according to the electric power load capacity of the area indicated in the prompt information.
For ease of understanding, the technical solution of embodiments of the present disclosure is described in detail by taking a resource processing scenario as an example.
In step 901, a parameter optimization request for a browsing parameter corresponding to a browsing operation initiated by the target user is received.
In step 902, a parameter sampling algorithm matching a browsing objective of the target user is determined.
In step 903, in response to a sample acquisition request initiated by the target user for the browsing parameter, the parameter sampling algorithm is invoked to generate a test sample for the browsing parameter.
In step 904, a simulation result for the test sample is determined based on a preset objective function.
In step 905, the simulation result corresponding to the test sample for the browsing parameter is outputted to the target user, for the target user to configure the browsing parameter in accordance with the test sample.
In the field of e-commerce, the parameter optimization problem is also involved. Take the common product recommendation as an example, due to different browsing characteristics of users, such as the consumption habits, fields of interest, and historical browsing behaviors, the content or products recommended for users are also different. In practical application, in order to improve the user's click-through rate, the user's browsing characteristics such as consumption habits and fields of interest can be parameterized to generate different browsing parameters, and through the setting of the plurality of browsing parameters, accurate analysis of the characteristics of the user's click-through objective can be achieved, to find the target products with higher user attention. The solution for performing parameter sampling of a plurality of browsing parameters and conducting parameter trials to determine the click-through rate of the user can be applied to the technical solution of embodiments of the present disclosure to improve the efficiency of trials.
The request detection module 1001 is configured to detect a parameter optimization request initiated by a target user and determine a parameter sampling algorithm matching the target user.
The first response module 1002 is configured to invoke, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a test sample.
The first simulation module 1003 is configured to determine a simulation result for the test sample based on a preset objective function.
The first output module 1004 is configured to output the simulation result for the test sample for the target user.
In some embodiments, the first simulation module may include a first sending unit, and a first receiving unit.
The first sending unit is configured to send the test sample to a user end of the target user, for the user end to input the test sample to the preset objective function to calculate and obtain the simulation result corresponding to the test sample.
The first receiving unit is configured to receive the simulation result corresponding to the test sample sent by the user end.
In some other embodiments, the first simulation module may include a first calculation unit configured to input the test sample to the preset objective function to calculate and obtain the simulation result for the test sample.
In some embodiments, the apparatus may further include a node determination unit configured to determine a target node for starting the parameter sampling algorithm.
The first response module may include a first response unit and a sample acquisition unit.
The first response unit is configured to send, in response to the sample acquisition request, the sample acquisition request to the target node starting the parameter sampling algorithm.
The sample acquisition unit is configured to acquire, through the target node, the test sample generated by the parameter sampling algorithm.
In a possible design, the first calculation unit may include a module establishing unit and a module calculation unit.
The module establishing unit is configured to establish a simulation module for the objective function.
The module calculation unit is configured to input the test sample to the simulation module to calculate and obtain the simulation result for the test sample.
In some other embodiments, the apparatus may further include an identifier receiving module and an identifier marking module.
The identifier receiving module is configured to receive a simulation trial identifier sent by the target user.
The identifier marking module is configured to store the simulation result for the test sample for the target user based on the simulation trial identifier.
In some embodiments, the request detection module may include a request receiving unit and a first storage unit.
The request receiving unit is configured to receive the parameter optimization request sent by the target user, in which the parameter optimization request includes request service information for the parameter sampling algorithm.
The first storage unit is configured to determine, upon detecting the presence of a storage operation for the request service information in the data storage component, the parameter sampling algorithm for the target user to request service in the request service information.
As a possible implementation, the apparatus may further include a first verification unit and a second storage unit.
The first verification unit is configured to verify the parameter optimization request of the target user based on the verification information in the request service information to obtain a verification result.
The second storage unit is configured to store the request service information to the data storage component in response to the verification result being a successful verification.
In some embodiments, the first storage unit may be used specifically configured to acquire, via a resource controller, a storage notification message initiated by the data storage component when storing the request service information and determine, based on the storage notification message, the parameter sampling algorithm for the target user to request service in the request service information.
In some embodiments, the apparatus may further include a convergence judgment module, a first processing module, and a second processing module.
The convergence judgment module is configured to acquire a judgment result as to whether the test sample satisfies a preset convergence condition.
The first processing module is configured to determine a target sample based on the test sample in response to the judgment result being that the test sample satisfies the convergence condition.
The second processing module is configured to redirect to the first response module to continue execution in response to the judgment result being that the test sample does not satisfy the convergence condition.
In one possible design, the first response module may include a second response unit configured to invoke, in response to the sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a plurality of test samples in a batch processing manner.
The first simulation module may include a first simulation unit configured to determine simulation results separately corresponding to the plurality of test samples based on the preset objective function.
The first output module may include a first output unit configured to output the simulation results separately corresponding to the plurality of test samples for the target user.
In yet another possible design, the first simulation module may include a simulation establishing unit, a simulation matching module, and a second simulation module.
The simulation establishing unit is configured to establish a plurality of simulation modules for the objective function.
The simulation matching module is configured to match target simulation modules separately for the plurality of test samples from the plurality of simulation modules.
The second simulation module is configured to input any test sample to the target simulation module corresponding to that test sample to calculate and obtain a simulation result for the test sample to obtain the simulation results separately corresponding to the plurality of test samples.
Further, optionally, the first simulation unit may specifically be configured to send the plurality of test samples to the user end of the target user for the user end to input the plurality of test samples separately to the preset objective function to calculate and obtain the simulation results separately corresponding to the plurality of test samples, and receive the simulation results separately corresponding to the plurality of test samples that are sent by the user end.
In some embodiments, the request detection module may include a second receiving unit and a first determination unit.
The second receiving unit is configured to receive a parameter optimization request initiated by a target user for a parameter to be processed of a target resource.
The first determination unit is configured to determine a parameter sampling algorithm matching a processing objective of the target resource.
The first response module may include a first request response unit configured to invoke, in response to a sample acquisition request initiated by the target user for the parameter to be processed, the parameter sampling algorithm to generate a test sample for the parameter to be processed.
The first output module may include a second output unit configured to output a simulation result corresponding to the test sample for the parameter to be processed to the target user, for the target user to process the target resource in accordance with the test sample.
In some other embodiments, the request detection module may include a third receiving unit and a second determination unit.
The third receiving unit is configured to receive a parameter optimization request for a browsing parameter generated for a browsing operation initiated by the target user.
The second determination unit is configured to determine a parameter sampling algorithm matching a browsing objective of the target user.
The first response module may include a second request response unit configured to invoke, in response to a sample acquisition request initiated by the target user for the browsing parameter, the parameter sampling algorithm to generate a test sample for the browsing parameter.
The first output module may include a third output unit configured to output the simulation result corresponding to the test sample for the browsing parameter to the target user, for the target user to configure the browsing parameter in accordance with the test sample.
In some embodiments, the objective function is a black box algorithm. The request detection module may specifically be configured to detect a black box optimization request initiated by the target user, and determine the parameter sampling algorithm matching the target user.
The information processing apparatus of
In a practical application, the information processing apparatus illustrated in FIG. 10 may be configured as a server. Referring to
The processing component 1102 may be configured to detect a parameter optimization request initiated by a target user and determining a parameter sampling algorithm matching the target user; invoke, in response to a sample acquisition request initiated by the target user, the parameter sampling algorithm to generate a test sample; determine a simulation result for the test sample based on a preset objective function; and output the simulation result for the test sample for the target user.
In addition, the processing component may also perform the information processing methods illustrated in any of the preceding embodiments, which will not be repeated herein for the sake of brevity of description. The server in embodiments of the present disclosure may be the service end of the preceding embodiments, and may perform interaction of data or information with the user device corresponding to the user end.
Here, the processing component 1102 may include one or more processors to execute the computer instructions to complete all or some of the steps in the methods described above. Of course, the processing component may also be one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, micro-controllers, micro-processors, or other electronic component implementations to perform the information processing methods described above.
The storage component 1101 is configured to store various types of data to support operations at the terminal. The storage component may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk.
Of course, the computing device can include other components, such as input/output interfaces, communication components, or the like. The input/output interfaces provide interfaces between the processing component and peripheral interface modules, which may be output devices, input devices, or the like. The communication component is configured to facilitate communication between the computing device and other devices by wired or wireless means, or the like.
In addition, embodiments of the present disclosure provide a computer-readable storage medium, wherein the storage medium may store one or more computer instructions, the one or more computer instructions, when executed, being used to implement any of the information processing methods in
The request initiation module 1201 is configured to detect a parameter optimization request initiated by a target user for a parameter sampling algorithm.
The first sending module 1202 is configured to send the parameter optimization request provided by the target user to a service end for the service end to determine a parameter sampling algorithm matching the target user.
The second sending module 1203 is configured to send a sample acquisition request initiated by the target user to the service end for the service end to invoke, in response to the sample acquisition request, the parameter sampling algorithm to generate a test sample.
The second simulation module 1204 is configured to determine a simulation result for the test sample based on a preset objective function.
The second output module 1205 is configured to output the simulation result for the test sample for the target user.
In some embodiments, the second simulation module may include a sample receiving unit and a sample simulation unit.
The sample receiving unit is configured to receive the test sample sent by the service end.
The sample simulation unit is configured to input the test sample to the objective function to calculate and obtain a calculation result corresponding to the test sample.
In some embodiments, the sample receiving unit may specifically be configured to receive a plurality of test samples sent by the service end, in which the plurality of test samples are generated by the service end by invoking the parameter sampling algorithm in a batch processing manner.
The sample simulation unit may specifically be configured to input the plurality of test samples separately to the objective function to calculate and obtain calculation results separately corresponding to the plurality of test samples.
In some other embodiments, the second simulation module may include a result receiving unit configured to receive the simulation result corresponding to the test sample sent by the service end. The simulation result is calculated and obtained by inputting the test sample to the preset objective function by the service end.
In some embodiments, the apparatus may further include a convergence judgment module, a first processing module and a second processing module.
The convergence judgment module is configured to acquire a judgment result as to whether the test sample satisfies a preset convergence condition.
The first processing module is configured to determine a target sample based on the test sample in response to the judgment result being that the test sample satisfies the convergence condition.
The second processing module is configured to redirect to the second sending module to continue execution in response to the judgment result being that the test sample does not satisfy the convergence condition.
The information processing apparatus of
In a practical application, the information processing apparatus illustrated in
The processing component 1302 may be configured to detect a parameter optimization request initiated by a target user for a parameter sampling algorithm, send the parameter optimization request to a service end for the service end to determine a parameter sampling algorithm matching the target user, send a sample acquisition request initiated by the target user to the service end for the service end to invoke, in response to the sample acquisition request, the parameter sampling algorithm to generate a test sample, determine a simulation result for the test sample based on a preset objective function, and output the simulation result for the test sample for the target user.
The user device illustrated in embodiments of the present disclosure can be specifically used as the user end in the preceding embodiments, and as to the specific processing methods and beneficial effects of each step in embodiments of the present disclosure, reference can be made to the relevant description of the user end in the preceding embodiments, which will not be repeated here.
In addition, the processing component may also perform the information processing methods illustrated in any of the preceding embodiments, which will not be repeated herein for the sake of brevity of description.
Here, the processing component 1302 may include one or more processors to execute the computer instructions to complete all or some of the steps in the methods described above. Of course, the processing component may also be one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, micro-controllers, micro-processors, or other electronic component implementations to perform the information processing methods described above.
The storage component 1301 is configured to store various types of data to support operations at the terminal. The storage component may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk.
Of course, the computing device can include other components, such as input/output interfaces, communication components, or the like. The input/output interfaces provide interfaces between the processing component and peripheral interface modules, which may be output devices, input devices, or the like. The communication component is configured to facilitate communication between the computing device and other devices by wired or wireless means, or the like.
In addition, embodiments of the present disclosure provide a computer-readable storage medium, wherein the storage medium may store one or more computer instructions, the one or more computer instructions, when executed, being used to implement the information processing methods illustrated in
The apparatus embodiments described above are only examples, where the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to a plurality of network units. Part or all of the modules may be selected according to actual needs to achieve the purpose of the solutions of the present embodiments. Those of ordinary skill in the art can understand and implement the solution of the present embodiments without creative effort.
From the description of the above implementations, those skilled in the art may clearly understand that the various implementations may be implemented by means of software plus a necessary general hardware platform, and may also be implemented by a combination of hardware and software. Based on such understanding, the above technical solutions essentially or the part contributing to the existing technology may be embodied in the form of a computer product. In the present disclosure, the form of a computer program product can be implemented on one or more computer-usable storage media (including, but not limited to, a magnetic disk memory, a CD-ROM, an optical memory, or the like) including computer usable program code.
The embodiments may further be described using the following clauses:
Finally, it should be noted that the above embodiments are only used for describing the technical solutions of the present disclosure, but are not intended to limit the present disclosure. Although the present disclosure is described in detail with reference to the foregoing embodiments, those of ordinary skill in the art may understand that modifications may still be made to the technical solutions described in the foregoing embodiments, or equivalent replacements may be made to some of the technical features. These modifications or replacements do not make the essence of the corresponding technical solutions depart from the spirit and scope of the technical solutions of the embodiments of the present disclosure.
| Number | Date | Country | Kind |
|---|---|---|---|
| 202110099097.6 | Jan 2021 | CN | national |
The present application claims priority to International Application No. PCT/CN2022/072913, filed Jan. 20, 2022, which claims priority to and the benefits of Chinese Patent Application No. 202110099097.6, filed on Jan. 25, 2021, both of which are incorporated herein by reference in their entireties.
| Number | Date | Country | |
|---|---|---|---|
| Parent | PCT/CN2022/072913 | Jan 2022 | US |
| Child | 18357442 | US |
