The present disclosure relates to a computer-implemented or hardware-implemented method for identification or separation of entities as well as to a computer program product, an apparatus, a transfer function unit and a system for entity identification or separation. More specifically, the disclosure relates to a computer-implemented or hardware-implemented method for identification or separation of entities as well as to a computer program product, an apparatus, a transfer function unit and a system for entity identification or separation as defined in the introductory parts of the independent claims.
Entity identification is known from prior art. One technology utilized for performing entity identification is neural networks. One type of neural network that can be utilized for entity identification is the Hopfield network. A Hopfield network is a form of recurrent artificial neural network. Hopfield networks serve as content-addressable (“associative”) memory systems with binary threshold nodes. A model sometimes utilized for entity identification is the Hodgkin-Huxley model, which describes how action potentials in neurons are initiated and propagated. Furthermore, the FitzHugh-Nagumo model, described at http://scholarpedia.org/article/FitzHugh-Nagumo_model is a model sometimes utilized to non-linearly modify a signal. However, the FitzHugh-Nagumo model is not normally utilized for entity identification.
However, existing neural network solutions can have inferior performance and/or low reliability for certain types of problems. Furthermore, the existing solutions take a considerable time to train and therefore may require a lot of computer power and/or energy, especially for training. Moreover, existing neural network solutions may require a lot of storage space. In addition, the output of the neural network or of a neural node may not have a sufficient dynamic range or the range for the output may not be dynamically adapted/adaptable. Furthermore, a system comprising a neural network or neural nodes may be very complex. Moreover, the input sensitivity may not be adaptable/variable. In addition, simultaneous identification of several different dynamic modes in the input may not be possible.
US 2008/0258767 A1 discloses computational nodes and computational-node networks that include dynamical-nanodevice connections. Furthermore, US 2008/0258767 A1 discloses that a node comprises a state machine. However, the state machine is controlled by a global clock, thus the state of every node is dependent on the global clock and the state machine is not independent.
Therefore, there is a need for alternative approaches of entity identification or separation. Preferably, such approaches provide or enable one or more of improved performance, higher reliability, increased efficiency, faster training, use of less computer power, use of less training data, use of less storage space, less complexity, provision of a wider dynamic range, provision of a (more) adaptable dynamic range for the output, provision of a more adaptable/variable input sensitivity, identification of several different dynamic modes in the input simultaneously and/or use of less energy.
An object of the present disclosure is to mitigate, alleviate or eliminate one or more of the above-identified deficiencies and disadvantages in the prior art and solve at least the above-mentioned problem.
According to a first aspect there is provided a computer-implemented or hardware-implemented method for identification or separation of entities. The method comprises receiving, by an input unit of a neural cell, a plurality of input signals from a plurality of sensors and/or from other neural cells. Furthermore, the method comprises scaling, by scaling unit of the neural cell, each of the plurality of input signals with a respective weight to obtain weighted input signals. Moreover, the method comprises calculating, by a summing unit of the neural cell, a sum of the weighted input signals to obtain a sum signal. The method comprises processing the sum signal, by a first processing unit of the neural cell, to obtain a first additional input signal. Furthermore, the method comprises amplifying the sum signal, by an amplifier of the neural cell, to obtain an amplified sum signal. Moreover, the method comprises adding, by an addition unit of the neural cell, the first additional input signal to the amplified sum signal to obtain an activity potential signal. The method comprises utilizing, by an output unit of the neural cell, the activity potential signal as a third additional input signal to the first processing unit of the neural cell and as an output signal for the neural cell to identify or separate entities. By utilizing the activity potential signal as the output signal for the neural cell, the range of the output signal can be dynamically adapted, thereby automatically providing a wide or wider dynamic range of the output, and thereby more accurately and/or efficiently identify or separate different entities, such as phonemes.
According to some embodiments, the method comprises transforming the first additional input signal, by a second processing unit of the neural cell, to obtain a second additional input signal. According to some embodiments, the step of adding further comprises adding, by the neural cell, the second additional input signal to the amplified sum signal to obtain the activity potential signal.
According to some embodiments, processing the sum signal, by a first processing unit of the neural cell, to obtain a first additional input signal comprises: checking whether the sum signal is positive or negative; if the sum signal is negative, feeding the sum signal to a first accumulator, thereby charging the first accumulator; if the sum signal is positive or zero, feeding the sum signal to a discharge unit connected to the first accumulator; and utilizing an output of the discharge unit as the first additional input signal.
According to some embodiments, utilizing the activity potential signal as a third additional input signal to the first processing unit of the neural cell comprises: checking whether the activity potential signal is positive or negative; if the activity potential signal is negative, feeding the activity potential signal to the first accumulator, thereby charging the first accumulator; if the activity potential signal is positive or zero, feeding the activity potential signal to the discharge unit.
By implementing the neural cell (or the transfer function unit thereof) with an accumulator and a discharge unit, a highly non-linear input-output relationship which varies over time (depending on previous inputs) is achieved, thereby improving/increasing separability and/or the ability to identify an entity (e.g., as the resolution is improved). Furthermore, by implementing each neural cell (or the transfer function unit of each neural cell) of a network with an accumulator (and a discharge unit), each neural cell is provided with an intrinsic memory function (i.e., the accumulator carries cell memory properties), which is independent of other neural cell's intrinsic memories and independent of global control signals, such as global clock inputs, thus providing a more flexible system/network, which has a higher capacity to separate a higher number of entities or more accurately identifies entities. Moreover, by providing each neural cell of a network with an independent memory, the complexity of the system is reduced, e.g., since there is no need for an external clock, and/or a wider dynamic range, a greater diversity, learning with fewer resources and/or more efficient (independent) learning is achieved.
According to some embodiments, transforming the first additional input signal, by a second processing unit of the neural cell, to obtain a second additional input signal comprises: providing the first additional input signal to a second accumulator; low pass filtering an output of the second accumulator with a low pass filter to create a low-pass filtered version of the output of the second accumulator; comparing, with a comparator, the output of the second accumulator with the low-pass filtered version to create a negative difference signal; amplifying the negative difference signal with an amplifier to obtain a second additional input signal. According to some embodiments, the amplified negative difference signal is low pass or high pass filtered to obtain the second additional input signal.
According to some embodiments, the method comprises: receiving, at a compartment of the neural cell, a plurality of compartment input signals from a plurality of sensors and/or from other neural cells; scaling, by the compartment, each of the plurality of compartment input signals with a respective weight to obtain weighted compartment input signals; calculating, by the compartment, a sum of the weighted compartment input signals to obtain a compartment sum signal; processing the compartment sum signal, by a first compartment processing unit, to obtain a first additional compartment input signal; optionally transforming the first additional compartment input signal, by a second compartment processing unit, to obtain a second compartment additional input signal; amplifying the sum signal, by an amplifier of the compartment, to obtain an amplified compartment sum signal; adding, by the compartment, the first and optionally the second additional compartment input signals to the amplified compartment sum signal to obtain a compartment activity potential signal; and utilizing the compartment activity potential signal as a third additional compartment input signal to the first compartment processing unit and as a compartment output signal to adjust the sum signal based on a transfer function.
According to some embodiments, the plurality of input signals changes dynamically over time, and the activity potential signal is utilized to identify an entity, such as an object or a feature of an object, by comparing over a time period the activity potential signal to known activity potential signals associated with known entities and identifying the entity as the known entity which is associated with the known activity potential signal which is most similar to the activity potential signal.
According to some embodiments, the plurality of input signals changes dynamically over time, and the variation of the activity potential signal over time is measured by a post-processing unit, and the post-processing unit is configured to compare the measured variation to known measurable characteristics of entities, such as features of an object, comprised in a list associated with the post-processing unit and the post-processing unit is configured to identify an entity based on the comparison.
According to a second aspect there is provided a computer program product comprising a non-transitory computer readable medium, having thereon a computer program comprising program instructions, the computer program being loadable into a data processing unit and configured to cause execution of the method of the first aspect or any of the above mentioned embodiments when the computer program is run by the data processing unit.
According to a third aspect there is provided an apparatus for identification or separation of entities, comprising controlling circuitry configured to cause: reception of a plurality of input signals from a plurality of sensors and/or from other neural cells; scaling of each of the plurality of input signals with a respective weight to obtain weighted input signals; calculation of a sum of the weighted input signals to obtain a sum signal; processing of the sum signal to obtain a first additional input signal; amplification of the sum signal to obtain an amplified sum signal; optionally transformation of the first additional input signal to obtain a second additional input signal; addition of the first additional input signal, and optionally of the second additional input signal, to the amplified sum signal to obtain an activity potential signal; and utilization of the activity potential signal as a third additional input signal to a first processing unit and as an output signal to identify or separate entities.
According to some embodiments, the controlling circuitry is configured to cause processing of the sum signal, by a first processing unit of the neural cell, to obtain a first additional input signal by causing: checking of whether the sum signal is positive or negative; if the sum signal is negative, feeding of the sum signal to a first accumulator, thereby charging the first accumulator; if the sum signal is positive or zero, feeding of the sum signal to a discharge unit connected to the first accumulator; and utilization of an output of the discharge unit as the first additional input signal.
According to some embodiments, the controlling circuitry is configured to cause utilization of the activity potential signal as a third additional input signal to the first processing unit of the neural cell by causing: checking of whether the activity potential signal is positive or negative; if the activity potential signal is negative, feeding of the activity potential signal to the first accumulator, thereby charging the first accumulator; and if the activity potential signal is positive or zero, feeding of the activity potential signal to the discharge unit.
According to a fourth aspect there is provided a transfer function unit for adjusting the dynamics of a signal, the transfer function unit comprising: a reception unit configured to receive an input signal; an amplifier configured to amplify the input signal to obtain an amplified input signal; a first processing unit preferably comprising a first checking unit, the first checking unit is configured to check whether the input signal is positive or negative, configured to feed the input signal to a first accumulator if the input signal is negative and configured to feed the input signal to a discharge unit connected to the first accumulator if the input signal is positive or zero, and the first processing unit is configured to process the input signal to obtain a first additional input signal by utilizing an output of the discharge unit as the first additional input signal; an addition unit configured to add the first additional input signal to the amplified input signal to obtain an activity potential signal; and an output unit configured to provide the activity potential signal as a third additional input signal to the first processing unit and as an output signal for the neural cell, the dynamics of the output signal being different from the dynamics of the input signal.
According to some embodiments, the first processing unit comprises a second checking unit, the second checking unit is configured to check whether the activity potential signal is positive or negative; configured to feed the activity potential signal to the first accumulator if the activity potential signal is negative; and configured to feed the activity potential signal to the discharge unit if the activity potential signal is positive or zero.
According to a fifth aspect there is provided a system for identifying or separating entities comprising a plurality of neural cells. Each neural cell comprises: an input unit, configured to receive a plurality of input signals from a plurality of sensors and/or from other neural cells; a scaling unit, configured to scale each of the plurality of input signals with a respective weight to obtain weighted input signals; a summing unit, configured to calculate a sum of the weighted input signals to obtain a sum signal; and the transfer function unit of the fourth aspect. The sum signal is utilized as the input signal for the transfer function unit. The output signals of the transfer function units of the plurality of neural cells are utilized to identify or separate entities.
According to some embodiments, the first processing unit comprises a second checking unit, the second checking unit is configured to check whether the activity potential signal is positive or negative; configured to feed the activity potential signal to the first accumulator if the activity potential signal is negative; and configured to feed the activity potential signal to the discharge unit if the activity potential signal is positive or zero.
According to some embodiments, the system comprises a classifier comprising a list of known entities, such as objects, wherein each known entity is mapped to a respective (known) distribution of activity potential signals of each neural cell and the classifier is configured to receive the activity potential signal of each neural cell, configured to compare the activity potential signal of each neural cell to the distributions of activity potential signals of the known entities over a time period, and configured to identify the entity as one of the entities of the list based on the comparison.
According to some embodiments, the list is implemented as a look-up table, LUT.
According to some embodiments, the plurality of input signals changes dynamically over time and follows a sensor input trajectory.
According to some embodiments, the plurality of input signals are pixel values, such as intensity, of images captured by a camera and wherein the activity potential signal of each neural cell is further utilized to control a position of the camera by rotational and/or translational movement of the camera, thereby controlling the sensor input trajectory and wherein the entity identified is an object or a feature of an object present in one or more images of the captured images.
According to some embodiments, the plurality of sensors are touch sensors and the input from each of the plurality of sensors is a touch event signal with a force dependent value and the activity potential signal of each neural cell is utilized to identify the sensor input trajectory as a new contact event, the end of a contact event, a gesture or as an applied pressure.
According to some embodiments, each sensor of the plurality of sensors is associated with a different frequency band of an audio signal, wherein each sensor reports an energy present in the associated frequency band, and wherein the combined input from a plurality of such sensors follows a sensor input trajectory, and wherein the activity potential signal of each neural cell is utilized to identify a speaker and/or a spoken letter, a syllable, a phoneme, a word or a phrase present in the audio signal.
According to some embodiments, the plurality of sensors comprise a plurality of sensors related to a speaker, such as microphones, and wherein the output signal for the neural cell is utilized to identify or separate one or more speakers.
Effects and features of the second, third, fourth and fifth aspects are to a large extent analogous to those described above in connection with the first aspect and vice versa. Embodiments mentioned in relation to the first aspect are largely or fully compatible with the second, third, fourth and fifth aspects and vice versa.
An advantage of some embodiments is that the range of the output signal can be dynamically adapted.
Another advantage of some embodiments is that a wide or wider dynamic range of the output can be automatically provided.
Yet another advantage of some embodiments is that different entities, such as dynamic entities, e.g., phonemes, can be more accurately and/or efficiently identified or separated. A dynamic entity can exist in any sensing system, provided that it has a plurality of sensors whose activity level will differ in time from each other, when applied to the same measurement situation. A dynamic entity is here defined as a spatiotemporal pattern of activity levels across the plurality of sensors. The statistically recurring spatiotemporal patterns of sensor activity levels can correspond to a set of such dynamic entities that are useful to identify the structure of the time-evolving sensor data.
Another advantage of some embodiments is that a learning signal that is formed on basis of such dynamic entities can be present in a single node. Each node can then learn to identify a subset of dynamic entities. In a system of nodes, each node can learn to efficiently identify a potentially unique subset of entities, such as dynamic entities. A large number of nodes can then be used to identify a large number of entities, such as dynamic entities, potentially providing the system with a greater maximal performance.
An advantage of some embodiments is that a less complex system is obtained, e.g., since every component has an equivalent basic electrical/electronic component, and the entire system can be constructed using a limited set of standard electronic components.
The present disclosure will become apparent from the detailed description given below. The detailed description and specific examples disclose preferred embodiments of the disclosure by way of illustration only. Those skilled in the art understand from guidance in the detailed description that changes and modifications may be made within the scope of the disclosure.
Hence, it is to be understood that the herein disclosed disclosure is not limited to the particular component parts of the device described or steps of the methods described since such apparatus and method may vary. It is also to be understood that the terminology used herein is for purpose of describing particular embodiments only and is not intended to be limiting. It should be noted that, as used in the specification and the appended claim, the articles “a”, “an”, “the”, and “said” are intended to mean that there are one or more of the elements unless the context explicitly dictates otherwise. Thus, for example, reference to “a unit” or “the unit” may include several devices, and the like. Furthermore, the words “comprising”, “including”, “containing” and similar wordings does not exclude other elements or steps.
The above objects, as well as additional objects, features, and advantages of the present disclosure, will be more fully appreciated by reference to the following illustrative and non-limiting detailed description of example embodiments of the present disclosure, when taken in conjunction with the accompanying drawings.
The present disclosure will now be described with reference to the accompanying drawings, in which preferred example embodiments of the disclosure are shown. The disclosure may, however, be embodied in other forms and should not be construed as limited to the herein disclosed embodiments. The disclosed embodiments are provided to fully convey the scope of the disclosure to the skilled person.
The term “measurable” is to be interpreted as something that can be measured or detected, i.e., is detectable. The terms “measure” and “sense” are to be interpreted as synonyms.
The term entity is to be interpreted as an entity, such as physical entity or a more abstract entity, such as a financial entity, e.g., one or more financial data sets. The term “physical entity” is to be interpreted as an entity that has physical existence, such as an object, a feature (of an object), a gesture, an applied pressure, a speaker, a spoken letter, a syllable, a phoneme, a word, or a phrase.
The term “node” or “cell” may be a neuron (of a neural network) or another processing element.
Separation refers to the process of distinguishing an entity from another entity, e.g., distinguishing a phoneme from another phoneme.
Identification refers to the process of identification, wherein a certain entity is distinguished from other entities and then classified as a known entity, e.g., by a classifier utilizing a list of known entities. Generally, the identification is a biometrics identification/authentication. One example is speaker recognition/identification, i.e., voice biometry. However, the identification may instead be image analysis, such as dynamic image analysis, e.g., inter image analysis and/or prediction, i.e., analysis and/or prediction between different (subsequent) images.
The term “time-continuous data” or “time-continuous signal” (or “continuous-time data” or “continuous-time signal”) is to be interpreted as a signal of continuous amplitude and time, such as an analog signal.
In the following, embodiments will be described where
According to some embodiments, a computer program product comprises a non-transitory computer readable medium 1100 such as, for example a universal serial bus (USB) memory, a plug-in card, an embedded drive, a digital versatile disc (DVD) or a read only memory (ROM).
Furthermore, the controlling circuitry is configured to cause amplification 1250 of the sum signal 140 to obtain an amplified sum signal 144. To this end, the controlling circuitry may be associated with (e.g., operatively connectable, or connected, to) an amplifying unit (e.g., an amplifier 141 of the neural cell 100 or amplification circuitry). Optionally, the controlling circuitry is configured to cause transformation 1251 of the first additional input signal 150 to obtain a second additional input signal 160. To this end, the controlling circuitry may be associated with (e.g., operatively connectable, or connected, to) a second processing unit (second processing unit 190 of the neural cell 100 or a second processor). The controlling circuitry is configured to cause addition 1260 of the first additional input signal 150, and optionally the second additional input signal 160, to the amplified sum signal 144 to obtain an activity potential signal 170. To this end, the controlling circuitry may be associated with (e.g., operatively connectable, or connected, to) an addition unit 192 (an adder or addition circuitry). Furthermore, the controlling circuitry is configured to cause utilization 1270 of the activity potential signal 170 as a third additional input signal to the first processing unit 180 and as an output signal to identify or separate entities (or measurable characteristics thereof). To this end, the controlling circuitry may be associated with (e.g., operatively connectable, or connected, to) an output unit (output circuitry or output module).
Furthermore, in some embodiments, the step of utilization 1270 of the activity potential signal 170 as a third additional input signal to the first processing unit 180 of the neural cell 100 comprises checking 1272 of whether the activity potential signal 170 is positive or negative. If the activity potential signal 170 is negative, the step 1270 comprises feeding 1274 of the activity potential signal 170 to the first accumulator 304, thereby charging the first accumulator 304. If the activity potential signal 170 is positive or zero, the step 1270 comprises feeding 1276 of the activity potential signal 170 to the discharge unit 305. To this end, the controlling circuitry may be associated with (e.g., operatively connectable, or connected, to) a second checking unit (second checking circuitry or a second checker).
Each neural cell 100a, 100b, . . . , 100x comprises: an input unit, configured to receive a plurality of input signals 110a, 110b, . . . , 110x from a plurality of sensors and/or from other neural cells; a scaling unit, configured to scale each of the plurality of input signals 110a, 110b, . . . , 110x with a respective weight 120a, 120b, . . . , 120x to obtain weighted input signals 130a, 130b, . . . , 130x; a summing unit 135, configured to calculate a sum of the weighted input signals 130a, 130b, . . . , 130x to obtain a sum signal 140. Furthermore, each neural cell 100a, 100b, . . . , 100x comprises the transfer function unit 145 described in connection with
As mentioned in connection with
In some embodiments, the plurality of input signals 110a, 110b, . . . , 110x are pixel values, such as intensity or color, of images captured by a camera. If the camera moves across a visual field, then specific entities can generate specific sensor input trajectories. Statistically dominant such sensor input trajectories can be used to describe the dynamic entities existing in the visual scene, possibly as a function of the parameters of the camera movement. The entity identified is an object, such as a tree, a house, or a person, or a feature of an object, such as the distance between the eyes of a person, present in at least one image of the captured images. The system 1300 may comprise or be connected/connectable to the camera and means, such as one or more electrical motors, for rotational and/or translational movement of the camera.
In some embodiments, the plurality of sensors are touch sensors and the plurality of input signals 110a, 110b, . . . , 110x from each of the plurality of sensors are touch event signals with force dependent values, e.g., values from 0 to 1. In some embodiments, the force dependent values are compared to a threshold to create a binary value, e.g., 0 or 1, for the plurality of input signals 110a, 110b, . . . , 110x. The activity potential signal 170 of each neural cell 100 is utilized to identify the sensor input trajectory as a new contact event, the end of a contact event, a gesture or as an applied pressure. In some embodiments, each sensor of the plurality of sensors is associated with a different frequency band of an audio signal. Each sensor reports an energy present in the associated frequency band. The combined input from a plurality of the sensors follows a sensor input trajectory. The activity potential signal 170 of each neural cell 100 is utilized to identify a speaker and/or a spoken letter, syllable, phoneme, word, or phrase present in the audio signal. In some embodiments, the plurality of sensors comprise a plurality of sensors related to a speaker, such as microphones. The output signal for the neural cell 100 is utilized to identify or separate one or more speakers.
1. A computer-implemented or hardware-implemented method (200) for identification or separation of entities, comprising:
receiving (210), at a neural cell (100), a plurality of input signals (110a, 110b, . . . , 110x) from a plurality of sensors and/or from other neural cells;
scaling (220), by the neural cell (100), each of the plurality of input signals (110a, 110b, . . . , 110x) with a respective weight (120a, 120b, . . . , 120x) to obtain weighted input signals (130a, 130b, . . . , 130x);
calculating (230), by the neural cell (100), a sum of the weighted input signals (130a, 130b, . . . , 130x) to obtain a sum signal (140);
processing (240) the sum signal (140), by a first processing unit (180) of the neural cell (100), to obtain a first additional input signal (150);
amplifying (250) the sum signal (140), by an amplifier (141) of the neural cell (100), to obtain an amplified sum signal (144);
adding (260), by the neural cell (100), the first additional input signal (150) to the amplified sum signal (144) to obtain an activity potential signal (170); and
utilizing (270) the activity potential signal (170) as a third additional input signal to the first processing unit (180) of the neural cell (100) and as an output signal for the neural cell (100) to identify or separate entities.
2. The method of example 1, further comprising:
transforming (251) the first additional input signal (150), by a second processing unit (190) of the neural cell (100), to obtain a second additional input signal (160); and
wherein adding (260) further comprises adding, by the neural cell (100), the second additional input signal (160) to the amplified sum signal (144) to obtain the activity potential signal (170).
3. The method of any of examples 1-2, wherein processing (240) the sum signal (140), by a first processing unit (180) of the neural cell (100), to obtain a first additional input signal (150) comprises:
checking (242) whether the sum signal is positive or negative;
if the sum signal (140) is negative, feeding (244) the sum signal (140) to a first accumulator (304), thereby charging the first accumulator (304);
if the sum signal (140) is positive or zero, feeding (246) the sum signal (140) to a discharge unit (305) connected to the first accumulator (304); and
utilizing (248) an output of the discharge unit (305) as the first additional input signal (150); and/or
wherein utilizing (270) the activity potential signal (170) as a third additional input signal to the first processing unit (180) of the neural cell (100) comprises:
checking (272) whether the activity potential signal (170) is positive or negative;
if the activity potential signal (170) is negative, feeding (274) the activity potential signal (170) to the first accumulator (304), thereby charging the first accumulator (304);
if the activity potential signal (170) is positive or zero, feeding (276) the activity potential signal (170) to the discharge unit (305) and optionally
wherein transforming (251) the first additional input signal (150), by a second processing unit (190) of the neural cell (100), to obtain a second additional input signal (160) comprises:
providing (252) the first additional input signal (150) to a second accumulator (407);
low pass filtering (254) an output of the second accumulator (407) with a low pass filter (409) to create a low-pass filtered version of the output of the second accumulator (407);
comparing (256), with a comparator (412), the output of the second accumulator (407) with the low-pass filtered version to create a negative difference signal (410);
amplifying (258) the negative difference signal (410) with an amplifier (414), and optionally low pass or high pass filter (411) the amplified negative difference signal, to obtain a second additional input signal (160).
4. The method of any of examples 1-3, further comprising:
receiving (201), at a compartment (900) of the neural cell (100), a plurality of compartment input signals (910a, 910b, . . . , 910x) from a plurality of sensors and/or from other neural cells;
scaling (202), by the compartment (900), each of the plurality of compartment input signals (910a, 910b, . . . , 910x) with a respective weight (920a, 920b, . . . , 920x) to obtain weighted compartment input signals (930a, 930b, . . . , 930x);
calculating (203), by the compartment (900), a sum of the weighted compartment input signals (930a, 930b, . . . , 930x) to obtain a compartment sum signal (940);
processing (204) the compartment sum signal (940), by a first compartment processing unit (980), to obtain a first additional compartment input signal (950);
optionally transforming the first additional compartment input signal (950), by a second compartment processing unit (990), to obtain a second compartment additional input signal (960);
amplifying (205) the compartment sum signal (940), by an amplifier (941) of the compartment (900), to obtain an amplified compartment sum signal (944);
adding (206), by the compartment (900), the first and optionally the second additional compartment input signals (950, 960) to the amplified compartment sum signal (940) to obtain a compartment activity potential signal (970); and
utilizing (232) the compartment activity potential signal (970) as a third additional compartment input signal to the first compartment processing unit (980) and as a compartment output signal to adjust the sum signal (140) based on a transfer function.
5. The method of any of examples 1-4, further comprising adjusting (235), by the neural cell (100), the activity potential signal (170) based on a threshold function (142) and/or wherein each respective weight (120a, 120b, . . . , 120x) is updated based on a combination, such as a correlation, of the activity potential signal (170) and an input activity or a state of each respective weight (120a, 120b, . . . , 120x).
6. A computer program product comprising a non-transitory computer readable medium (1000), having thereon a computer program comprising program instructions, the computer program being loadable into a data processing unit (1020) and configured to cause execution of the method according to any of examples 1-5 when the computer program is run by the data processing unit (1020).
7. An apparatus for identification or separation of entities, comprising controlling circuitry configured to cause:
reception (1210) of a plurality of input signals (110a, 110b, . . . , 110x) from a plurality of sensors and/or from other neural cells;
scaling (1220) of each of the plurality of input signals (110a, 110b, . . . , 110x) with a respective weight (120a, 120b, . . . , 120x) to obtain weighted input signals (130a, 130b, . . . , 130x);
calculation (1230) of a sum of the weighted input signals (130a, 130b, . . . , 130x) to obtain a sum signal (140);
processing (1240) of the sum signal (140) to obtain a first additional input signal (150);
amplification (1250) of the sum signal (140) to obtain an amplified sum signal (144);
optionally transformation (1251) of the first additional input signal (150) to obtain a second additional input signal (160);
addition (1260) of the first additional input signal (150), and optionally of the second additional input signal (160), to the amplified sum signal (144), to obtain an activity potential signal (170); and
utilization (1270) of the activity potential signal (170) as a third additional input signal to the first processing unit (180) of the neural cell (100) and as an output signal to identify or separate entities.
8. A transfer function unit for adjusting the dynamics of a signal, the transfer function unit comprising:
a reception unit configured to receive an input signal (140);
an amplifier (141) configured to amplify the input signal (140) to obtain an amplified input signal (144);
a first processing unit (180) configured to process the input signal (140) to obtain a first additional input signal (150);
an addition unit (192) configured to add the first additional input signal (150) to the amplified input signal (144) to obtain an activity potential signal (170); and
an output unit configured to provide the activity potential signal (170) as a third additional input signal to the first processing unit (180) and as an output signal, the dynamics of the output signal being different from the dynamics of the input signal (140).
9. A system (1300) for identifying or separating entities comprising:
a plurality of neural cells (100a, 100b, . . . , 100x), each neural cell (100a, 100b, . . . , 100x) comprising: an input unit, configured to receive a plurality of input signals (110a, 110b, . . . , 110x) from a plurality of sensors and/or from other neural cells (100a, 100b, . . . , 100x);
a scaling unit, configured to scale each of the plurality of input signals (110a, 110b, . . . , 110x) with a respective weight (120a, 120b, . . . , 120x) to obtain weighted input signals (130a, 130b, . . . , 130x);
a summing unit (135), configured to calculate a sum of the weighted input signals (130a, 130b, . . . , 130x) to obtain a sum signal (140); and
the transfer function unit (145) of example 8; and
wherein the sum signal (140) is utilized as the input signal for the transfer function unit (145) and
wherein the output signals of the transfer function units (145) of the plurality of neural cells (100) are utilized to identify or separate entities.
10. The system of example 9, wherein the plurality of input signals (110a, 110b, . . . , 110x) changes dynamically over time and follows a sensor input trajectory, and
wherein the plurality of input signals (110a, 110b, . . . , 110x) are pixel values, such as intensity, of images captured by a camera and wherein the activity potential signal (170) of each neural cell (100) is further utilized to control a position of the camera by rotational and/or translational movement of the camera, thereby controlling the sensor input trajectory and wherein the entity identified is an object or a feature of an object present in one or more images of the captured images, or
wherein the plurality of sensors are touch sensors and the input from each of the plurality of sensors is a touch event signal with a force dependent value and wherein the activity potential signal (170) of each neural cell (100) is utilized to identify the sensor input trajectory as a new contact event, the end of a contact event, a gesture or as an applied pressure, or
wherein each sensor of the plurality of sensors is associated with a different frequency band of an audio signal, wherein each sensor reports an energy present in the associated frequency band, and wherein the combined input from a plurality of such sensors follows a sensor input trajectory, and wherein the activity potential signal (170) of each neural cell (100) is utilized to identify a speaker and/or a spoken letter, a syllable, a phoneme, a word or a phrase present in the audio signal or
wherein the plurality of sensors comprise a plurality of sensors related to a speaker, such as microphones, and wherein the output signal for the neural cell (100) is utilized to identify or separate one or more speakers.
The person skilled in the art realizes that the present disclosure is not limited to the preferred embodiments described above. The person skilled in the art further realizes that modifications and variations are possible within the scope of the appended claims. For example, other entities such as aroma or flavor may be identified or separated. Additionally, variations to the disclosed embodiments can be understood and effected by the skilled person in practicing the claimed disclosure, from a study of the drawings, the disclosure, and the appended claims.
Number | Date | Country | Kind |
---|---|---|---|
2151099-5 | Sep 2021 | SE | national |
Filing Document | Filing Date | Country | Kind |
---|---|---|---|
PCT/SE2022/050766 | 8/26/2022 | WO |