Patent Application
20210209412
Aspects of the present invention relate generally to machine learning and, more particularly, to using automated weak supervision to label training data that is used to train a machine learning model.
Machine learning is a form of artificial intelligence (AI) that enables a system to learn from data rather than through explicit programming. In machine learning, a machine learning model is built using algorithms that iteratively learn from training data. Training data can be described as data points that include patterns, which the resulting machine learning model should accurately predict.
An example training technique includes supervised learning, in which the training data is labeled, and the labeled training data is processed (e.g., using linear regression) to infer the machine learning model. Weak supervision is a branch of machine learning where noisy, limited, or imprecise sources are used to provide supervision signal for labeling large amounts of training data in a supervised learning setting.
In a first aspect of the invention, there is a computer-implemented method including: receiving, by a computing device, data comprising a labeled dataset and an unlabeled dataset; generating, by the computing device, a set of heuristics using the labeled dataset; generating, by the computing device, a vector of initial labels by labeling each point in the unlabeled dataset using the set of heuristics; generating, by the computing device, a refined set of heuristics using data-driven active learning; generating, by the computing device, a vector of training labels by automatically labeling each point in the unlabeled dataset using the refined set of heuristics; and outputting, by the computing device, the vector of training labels to a client device or a data repository.
In another aspect of the invention, there is a computer program product including one or more computer readable storage media having program instructions collectively stored on the one or more computer readable storage media. The program instructions are executable to: receive data comprising a labeled dataset and an unlabeled dataset; generate a set of heuristics using the labeled dataset; generate a vector of initial labels by labeling each point in the unlabeled dataset using the set of heuristics; generate a refined set of heuristics using data-driven active learning; generate a vector of training labels by automatically labeling each point in the unlabeled dataset using the refined set of heuristics; and output the vector of training labels to a client device or a data repository.
In another aspect of the invention, there is system including a processor, a computer readable memory, one or more computer readable storage media, and program instructions collectively stored on the one or more computer readable storage media. The program instructions are executable to: receive data comprising a labeled dataset and an unlabeled dataset; generate a set of heuristics using the labeled dataset; generate a vector of initial labels by labeling each point in the unlabeled dataset using the set of heuristics; generate a refined set of heuristics using data-driven active learning; generate a vector of training labels by automatically labeling each point in the unlabeled dataset using the refined set of heuristics; and output the vector of training labels to a client device or a data repository.
Aspects of the present invention are described in the detailed description which follows, in reference to the noted plurality of drawings by way of non-limiting examples of exemplary embodiments of the present invention.
Aspects of the present invention relate generally to machine learning and, more particularly, to using automated weak supervision to label training data that is used to train a machine learning model. In embodiments, a first phase of a method includes a system receiving a labeled dataset and an unlabeled dataset, automatically generating a set of heuristics from the labeled dataset, and using the set of heuristics to produce initial labels for the data in the unlabeled dataset. According to aspects of the invention, the system generates the set of heuristics automatically without input from a human user. In embodiments, after generating the set of heuristics and the initial labels, a second phase of the method includes designing a query strategy based on the data distribution and output of the first phase, and using the query strategy to prompt a user to input true labels for a small subset of the data in a data-driven active learning procedure. In embodiments, the output of the second phase is a set of refined heuristics that the system uses in a third phase to produce probabilistic training labels for the data in the unlabeled dataset. In this manner, implementations of the invention automate aspects of weak supervision to produce labels for training data for machine learning models.
Organizations in different domains are increasingly investing in machine learning to empower their data-driven decisions. However, one of the most tedious tasks in creating machine learning models is obtaining hand-labeled training data, especially with the new revolutionary advances that deep learning methods bring to the field of machine learning. Since such techniques require large training datasets, the cost of labeling these datasets has become a significant expense for businesses and large organizations. In real-world settings, domain experience is usually required to accomplish, or at least supervise such labeling processes; this makes the process of obtaining large-scale hand-labeled training data prohibitively expensive.
Aspects of the present invention address these issues by providing a framework for generating high-quality labeled datasets at scale. In an embodiment, a method includes an iterative process to automatically generate high accuracy heuristics to assign initial labels to unlabeled data. In this embodiment, the method then applies a data-driven active learning process to further enhance the quality of the generated heuristics. In this embodiment, the method includes learning the active learning strategy while considering the modeled accuracies of the produced heuristics and the noise in the generated labels. In this embodiment, the method includes applying the learned strategy to economically engage the user and enhance the quality of the generated labels. In this manner, implementations of the invention are usable to provide labels for unlabeled data, which can then be used to train a machine learning model in a supervised learning context.
According to aspects of the invention, there is a computer-implemented process for generating training datasets, the computer-implemented process comprising: in response to receiving a set of labeled data, generating a set of heuristics and a set of generated weak labels using a first iterative process including creating, testing, and ranking heuristics in each, and every, iteration that only exceed a predetermined level of accuracy of heuristics; analyzing disagreements between the heuristics generated to model associated accuracies; applying a data-driven automated learning process to analyze the generated weak labels and modeled accuracies of the heuristics generated to identify only possible points to provide true labels; prompting a user to provide true labels only for the possible points; in response to receiving the true labels from the user, refining a set of initial labels generated by the heuristics, using a second iterative process to create refined labels; and in response to an examination of the refined labels meeting a predetermined threshold, creating a set of probabilistic labels for training a downstream classifier.
Implementations of the invention improve the performance of a computer system that is used to label data for use in training machine learning models. The inventors evaluated a framework in accordance with aspects of the invention by comparing its performance with other weak supervision techniques such as data programming and automated weak supervision, along with active learning strategies. The empirical results show that the framework in accordance with aspects of the invention can significantly enhance the learned accuracy of the generated heuristics by up 44%, while producing high coverage labels for up to 91% of the unlabeled dataset. Also, comparing to the weak supervision techniques, the results show that the framework in accordance with aspects of the invention improves the quality of the generated labels by 28% on average. As well, the framework in accordance with aspects of the invention can reduce the annotation effort by up to 53% when compared to the baseline active learning strategies. Aspects of the invention also have a practical application of generating training data by applying labels to previously unlabeled data, which training data can then be used in training a machine learning model.
The present invention may be a system, a method, and/or a computer program product at any possible technical detail level of integration. The computer program product may include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present invention.
The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. The computer readable storage medium may be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of the computer readable storage medium includes the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as punch-cards or raised structures in a groove having instructions recorded thereon, and any suitable combination of the foregoing. A computer readable storage medium or media, as used herein, is not to be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media (e.g., light pulses passing through a fiber-optic cable), or electrical signals transmitted through a wire.
Computer readable program instructions described herein can be downloaded to respective computing/processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and/or a wireless network. The network may comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers. A network adapter card or network interface in each computing/processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium within the respective computing/processing device.
Computer readable program instructions for carrying out operations of the present invention may be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, configuration data for integrated circuitry, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++, or the like, and procedural programming languages, such as the “C” programming language or similar programming languages. The computer readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate arrays (FPGA), or programmable logic arrays (PLA) may execute the computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry, in order to perform aspects of the present invention.
Aspects of the present invention are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer readable program instructions.
These computer readable program instructions may be provided to a processor of a computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, a programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer readable storage medium having instructions stored therein comprises an article of manufacture including instructions which implement aspects of the function/act specified in the flowchart and/or block diagram block or blocks.
The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process, such that the instructions which execute on the computer, other programmable apparatus, or other device implement the functions/acts specified in the flowchart and/or block diagram block or blocks.
The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the blocks may occur out of the order noted in the Figures. For example, two blocks shown in succession may, in fact, be accomplished as one step, executed concurrently, substantially concurrently, in a partially or wholly temporally overlapping manner, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems that perform the specified functions or acts or carry out combinations of special purpose hardware and computer instructions.
Referring now to
In computer infrastructure 10 there is a computer system 12, which is operational with numerous other general purpose or special purpose computing system environments or configurations. Examples of well-known computing systems, environments, and/or configurations that may be suitable for use with computer system 12 include, but are not limited to, personal computer systems, server computer systems, thin clients, thick clients, hand-held or laptop devices, multiprocessor systems, microprocessor-based systems, set top boxes, programmable consumer electronics, network PCs, minicomputer systems, mainframe computer systems, and distributed cloud computing environments that include any of the above systems or devices, and the like.
Computer system 12 may be described in the general context of computer system executable instructions, such as program modules, being executed by a computer system. Generally, program modules may include routines, programs, objects, components, logic, data structures, and so on that perform particular tasks or implement particular abstract data types. Computer system 12 may be practiced in distributed cloud computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed cloud computing environment, program modules may be located in both local and remote computer system storage media including memory storage devices.
As shown in
Bus 18 represents one or more of any of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, and a processor or local bus using any of a variety of bus architectures. By way of example, and not limitation, such architectures include Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MCA) bus, Enhanced ISA (EISA) bus, Video Electronics Standards Association (VESA) local bus, and Peripheral Component Interconnects (PCI) bus.
Computer system 12 typically includes a variety of computer system readable media. Such media may be any available media that is accessible by computer system 12, and it includes both volatile and non-volatile media, removable and non-removable media.
System memory 28 can include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and/or cache memory 32. Computer system 12 may further include other removable/non-removable, volatile/non-volatile computer system storage media. By way of example only, storage system 34 can be provided for reading from and writing to a non-removable, non-volatile magnetic media (not shown and typically called a “hard drive”). Although not shown, a magnetic disk drive for reading from and writing to a removable, non-volatile magnetic disk (e.g., a “floppy disk”), and an optical disk drive for reading from or writing to a removable, non-volatile optical disk such as a CD-ROM, DVD-ROM or other optical media can be provided. In such instances, each can be connected to bus 18 by one or more data media interfaces. As will be further depicted and described below, memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to carry out the functions of embodiments of the invention.
Program/utility 40, having a set (at least one) of program modules 42, may be stored in memory 28 by way of example, and not limitation, as well as an operating system, one or more application programs, other program modules, and program data. Each of the operating system, one or more application programs, other program modules, and program data or some combination thereof, may include an implementation of a networking environment. Program modules 42 generally carry out the functions and/or methodologies of embodiments of the invention as described herein.
Computer system 12 may also communicate with one or more external devices 14 such as a keyboard, a pointing device, a display 24, etc.; one or more devices that enable a user to interact with computer system 12; and/or any devices (e.g., network card, modem, etc.) that enable computer system 12 to communicate with one or more other computing devices. Such communication can occur via Input/Output (I/O) interfaces 22. Still yet, computer system 12 can communicate with one or more networks such as a local area network (LAN), a general wide area network (WAN), and/or a public network (e.g., the Internet) via network adapter 20. As depicted, network adapter 20 communicates with the other components of computer system 12 via bus 18. It should be understood that although not shown, other hardware and/or software components could be used in conjunction with computer system 12. Examples, include, but are not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc.
In embodiments, the labeling server 210 comprises a heuristics generator module 211, a data-driven leaner module 212, and a probabilistic labels generator module 213, each of which may comprise one or more program modules such as program modules 42 described with respect to
In accordance with aspects of the invention, the heuristics generator module 211 is configured to automatically produce a set of heuristics using a labeled dataset and use the heuristics to automatically assign initial labels to data points in an unlabeled dataset. In one implementation, the heuristics generator module 211 receives the labeled dataset and the unlabeled dataset from a client device 220 via a network 230. For example, a user may use a client application 221 running on the client device 220 to upload one or more computer readable files containing the labeled dataset and the unlabeled dataset to the heuristics generator module 211. In another implementation, the heuristics generator module 211 obtains the labeled dataset and the unlabeled dataset from a data repository 240. For example, a user may use the client application 221 running on the client device 220 to designate one or more computer readable files that are stored in the data repository 240 and that contain the labeled dataset and the unlabeled dataset, and based on this designation the heuristics generator module 211 may obtain the designated one or more computer readable files from the data repository 240. The data repository 240 may be included in the labeling server 210 or may be connected to the labeling server 210 via the network 230.
In accordance with aspects of the invention, the data-driven leaner module 212 is configured to work with the outcomes of the heuristics generator module 211 to further examine the data and refine the initial labels. In embodiments, the data-driven leaner module 212 is configured to enhance the accuracy of the generated heuristics and increase the coverage of the generated training data. In this manner, the data-driven leaner module 212 economically engages a user to express their domain experience and uses their input in the refinement process.
In accordance with aspects of the invention, the probabilistic labels generator module 213 is configured to learn the accuracy of these labels and assign a single label for each data point in the unlabeled dataset. In this manner, the probabilistic labels generator module 213 generates training data (e.g., by applying a label to each data point in the unlabeled dataset) that can be used to train a machine learning model. In embodiments, the probabilistic labels generator module 213 stores the training data as structured data in one or more computer-readable files in the data repository 240.
In embodiments, at step 301 the system (e.g., the labeling server 210) exploits a small set of labeled data and automatically produces a set of heuristics to assign initial labels to a larger unlabeled dataset. In this phase, the heuristics generator module 211 applies an iterative process of creating, testing, and ranking heuristics in each, and every, iteration to only accommodate high-quality heuristics. At step 302, the data-driven leaner module 212 examines disagreements between these heuristics (from step 301) to model their accuracies. In order to enhance the quality of the generated labels, at step 303, the data-driven leaner module 212 improves the accuracies of the heuristics by applying a data-driven active learning (AL) process. According to aspects of the invention, during this data-driven AL process, the system examines the generated weak labels along with the modeled accuracies of the heuristics to help the learner decide on the points for which the user should provide true labels. In this manner, implementations of the invention aim to enhance the accuracy and the coverage of the training data while engaging the user in the loop (e.g., via the client device 220) to execute the enhancement process. In accordance with aspects of the invention, by incorporating the underlying data representation, the user is only queried at step 303 about a subset of the points that are expected to enhance the overall labeling quality. In this manner, the manual labeling of data points by a domain expert is minimized. The true labels provided by the users are used to refine the initial labels generated by the heuristics. As the figure shows, the refinement process can be repeated to further enhance the quality of the generated labels. At step 304, the probabilistic labels generator module 213 examines the refined labels and outputs a set of probabilistic labels that can be used to train any downstream classifier (e.g., machine learning model).
Referring now to
In one example, the heuristics generator module 211 creates a set of weak classifiers by diving DL into a training dataset and an evaluation dataset, and employing one or more classifier algorithms and an iterative process of all possible combinations of input features to determine classifiers that perform well when applied to the evaluation dataset. Classifier algorithms may include, but are not limited to, decision stump algorithms and random forest algorithms. More specifically, in one exemplary implementation, the heuristics generator module 211 uses an ensemble of decision stumps as the inner classification model to mimic the threshold-based heuristics that users usually write.
In embodiments, to create the final set of heuristics H, the heuristics generator module 211 follows the iterative process of defining the input (features) for the potential models, creating the models (heuristics), and evaluating their performance and coverage. After these steps, the heuristics generator module 211 ranks the heuristics generated by each, and every, iteration to decide upon which heuristic to add to the set H. In accordance with aspects of the invention, the heuristics generator module 211 automatically generates the set of heuristics H using the techniques described herein, and does not prompt a user for input about the heuristics (e.g., does not employ user input in automatically generating the set of heuristics H).
In embodiments, to combine the output of the heuristics and generate the vector V of initial labels, the heuristics generator module 211 employs a generative model to learn the accuracies of the heuristics in H and produce a single probabilistic label for each data point in the unlabeled dataset. In one example, the heuristics generator module 211 generates a matrix in which each row of the matrix corresponds to one data point of DU and each column of the matrix corresponds to one heuristic of the set of heuristics H. In this example, the heuristics generator module 211 uses a generative model to create the vector V from the matrix, wherein the vector V includes a single probabilistic label for each data point in the unlabeled dataset DU.
Still referring to
In embodiments, the data-driven learner module 212 performs two processes. First, the data-driven learner module 212 designs an active learning (AL) query strategy that fits the data distribution for a given problem, e.g., based on H and V. Second, the data-driven learner module 212 applies the query strategy as a data-driven AL process. By utilizing these processes in accordance with aspects of the invention, a portion of the low confidence labels in the initial vector V of probabilistic labels is replaced by true labels, and a refined heuristics matrix RH is generated, which is an improved version of H.
In embodiments, the low confidence points are originated when either the heuristics abstain from labeling or disagree on specific points. Therefore, the data-driven learner module 212 enhances the quality of the labels by trying to eliminate the abstaining effect and resolve the disagreements between the heuristics to increase their accuracies.
In embodiments, the data-driven learner module 212 performs the second phase by determining a confidence level of each label in V (e.g., either high confidence or low confidence), and selecting one or more of the low confidence labels to present to a user so that the user can provide input defining true labels for the data points having the low confidence labels. In embodiments, a high confidence label is one in which the number of heuristics (in the matrix) that agree on the label exceeds a predefined threshold number, and a low confidence label is one in which the number of heuristics (in the matrix) that agree on the label is less than the predefined threshold number. In one example, the data-driven learner module 212 trains a regression model using the data in V (e.g., as described above), and then uses the regression model as a query strategy to determine which low confidence labels in V to present to the user for manual labeling. In this manner, the active learning is data-driven because it is based on the regression model that is trained using the data (e.g., the data in V), as opposed to a query strategy in which data points are randomly chosen for manual labeling.
With continued reference to
As depicted in
At step 505, the system receives data comprising a labeled dataset and an unlabeled dataset. In embodiments, and as described with respect to
At step 510, the system automatically generating a set of heuristics using the labeled dataset. In embodiments, and as described with respect to
At step 515, the system generates a vector of initial labels by automatically labeling each point in the unlabeled dataset using the set of heuristics. In embodiments, and as described with respect to
At step 520, the system generates a refined set of heuristics using data-driven active learning. In embodiments, and as described with respect to
At step 525, the system generates a vector of training labels by automatically labeling each point in the unlabeled dataset using the refined set of heuristics. In embodiments, and as described with respect to
At step 530, the system determines whether the process has produced a satisfactory result. In embodiments, and as described with respect to
In embodiments, a service provider could offer to perform the processes described herein. In this case, the service provider can create, maintain, deploy, support, etc., the computer infrastructure that performs the process steps of the invention for one or more customers. These customers may be, for example, any business that uses technology. In return, the service provider can receive payment from the customer(s) under a subscription and/or fee agreement and/or the service provider can receive payment from the sale of advertising content to one or more third parties.
In still additional embodiments, the invention provides a computer-implemented method, via a network. In this case, a computer infrastructure, such as computer system 12 (
The descriptions of the various embodiments of the present invention have been presented for purposes of illustration, but are not intended to be exhaustive or limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein was chosen to best explain the principles of the embodiments, the practical application or technical improvement over technologies found in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.
