Abstract: Described are platforms, systems, and methods to combine counts of activity correlations over time with a link salience method to identify collections of digital devices in an automated environment to identify sub-systems comprised of portions of the overall environment. The platforms, systems, and methods detect activity in a plurality of data sources associated with an automation environment; determine correlation in the detected activity between two or more of the data sources; store records of determined correlation in the detected activity over time in a data storage system; apply a link salience algorithm to the stored records of determined correlation in the detected activity to determine a salience property; and identify one or more subsystems in the automation environment based on the salience property.

Abstract: Methods, apparatus, systems and articles of manufacture for distributed training of a neural network are disclosed. An example apparatus includes a neural network trainer to select a plurality of training data items from a training data set based on a toggle rate of each item in the training data set. A neural network parameter memory is to store neural network training parameters. A neural network processor is to generate training data results from distributed training over multiple nodes of the neural network using the selected training data items and the neural network training parameters. The neural network trainer is to synchronize the training data results and to update the neural network training parameters.

Abstract: A system, method, and computer-readable medium are disclosed for predicting a defect within a computer program comprising: accessing a code base of the computer program, the code base of the computer program comprising a plurality of computer program files; training the defect prediction system, the training including performing a historical analysis of defect occurrence patterns in the code base of the computer program; analyzing a commit of the computer program to identify a likelihood of defect occurrence within each of the plurality of files of the computer program; and, calculating a defect prediction metric for each of the plurality of files of the computer program, the defect prediction metric providing an objective measure of defect prediction for each of the plurality of files of the computer program.

Abstract: Provided are a system for managing a calculation processing graph of an artificial neural network and a method of managing a calculation processing graph by using the system. A system for managing a calculation processing graph of an artificial neural network run by a plurality of heterogeneous resources includes: a task manager configured to allocate the plurality of heterogeneous resources to a first subgraph and a second subgraph that are to be run, the first subgraph and the second subgraph being included in the calculation processing graph; a first compiler configured to compile the first subgraph to be executable on a first resource among the plurality of heterogeneous resources; and a second compiler configured to compile the second subgraph to be executable on a second resource among the plurality of heterogeneous resources, wherein the first subgraph and the second subgraph are respectively managed through separate calculation paths.

Abstract: Disclosed is a method for implementing an adaptive stochastic spiking neuron based on a ferroelectric field effect transistor, relating to the technical field of spiking neurons in neuromorphic computing. Hardware in the method includes a ferroelectric field effect transistor (fefet), an n-type mosfet, and an l-fefet formed by enhancing a polarization degradation characteristic of a ferroelectric material for the ferroelectric field-effect transistor, wherein a series structure of the fefet and the n-type mosfet adaptively modulates a voltage pulse signal transmitted from a synapse. The l-fefet has a gate terminal connected to a source terminal of the fefet to receive the modulated pulse signal, and simulates integration, leakage, and stochastic spike firing characteristics of a biological neuron, thereby implementing an advanced function of adaptive stochastic spike firing of the neuron.

Abstract: Provided are systems and methods for operating a neural network processor, wherein the processor includes an input selector circuit that can be configured to select the data that will be input into the processor's computational array. In various implementations, the selector circuit can determine, for a row of the array, whether the row input will be the output from a buffer memory or data that the input selector circuit has selected for a different row. The row can receive an input feature map from a set of input data or an input feature map that was selected for inputting into a different row, such that the input feature map is input into more than one row at a time. The selector circuit can also include a delay circuit, so that the duplicated input feature map can be input into the computational array later than the original input feature map.

Abstract: A method of reducing kernel computations; the method comprising ordering a plurality of kernel channels. A first of the ordered kernel channels is then convolved with input data to produce a convolution output, and it is determined whether to convolve one or more subsequent kernel channels of the ordered kernel channels. Determining whether to convolve subsequent kernel channels comprises considering a potential contribution of at least one of the one or more subsequent kernel channels in combination with the convolution output.

Abstract: There is provided a neural network information receiving method and system, and a sending method and system. The receiving method comprises: acquiring a reception initiation time for neuron information (S100); receiving rostral neuron information output by rostral neurons (S200); acquiring delay information of the rostral neuron information according to the reception initiation time, the rostral neuron information and a delay algorithm (S300); and determining composite information output by the rostral neurons according to the rostral neuron information and the delay information (S400). The receiving method and system and sending method and system take the delay information into consideration in the output information of neurons, such that the neuron information is able to carry more detailed time-domain information, thus augmenting operation modes of the neurons and enhancing the generalization ability of the whole network.

Abstract: An industrial automation component may receive a first set of data associated with the industrial automation component, such that the industrial automation component is associated with a first industrial automation system. The industrial automation component may then receive a second set of data associated with one or more other industrial automation components, such that the one or more other industrial automation components are associated with one or more other industrial automation systems. The industrial automation component may then identify one or more similar patterns in the first set of data and the second set of data and adjust one or more operations of the industrial automation component based on the similar patterns.

Abstract: A computer program predictor is described which has a processor configured to access a program attribute predictor; and a memory storing a search component configured to search a space of possible programs, to find a program which, given an input data instance and an output data instance, will compute the output data instance from the input data instance, the search being guided by attributes predicted by the attribute predictor given the input data instance and the output data instance.

Abstract: A wind power prediction method and system based on a deep deterministic policy gradient (DDPG) algorithm is provided and relates to the technical field of wind power prediction. The method uses multiple different prediction methods to build a combined prediction sub-model, and then uses a DDPG algorithm to maximize discount benefit by using an agent in the algorithm to interact with an external prediction environment for constant trial-and-error attempts. Finally, the agent has a capability of perceiving the external prediction environment, and a capability of reasonably and dynamically assigning weights to various prediction sub-models in a combined model, so as to achieve an accurate prediction.

Abstract: Optimized synapses for neuromorphic arrays are provided. In various embodiments, first and second single-transistor current sources are electrically coupled in series. The first single-transistor current source is electrically coupled to both a first control circuit and second control circuit, free of any intervening logic gate between the first single-transistor current source and either one of the control circuits. The second single-transistor current source is electrically coupled to both the first control circuit and the second control circuit, free of any intervening logic gate between the second single-transistor current source and either one of the control circuits. A capacitor is electrically coupled to the first and second single-transistor current sources. A read circuit is electrically coupled to the capacitor. The first and second single-transistor current sources are adapted to charge the capacitor only when concurrently receiving a control signal from both the first and second control circuits.

Abstract: This application relates to computing circuitry (200, 500, 600) for analogue computing. A plurality of current generators (201) are each configured to generate a defined current (ID1, ID2, . . . IDj) based on a respective input data value (D1, D2, . . . Dj). A memory array (202), having at least one set (204) of programmable-resistance memory cells (203), is arranged to receive the defined currents from each of the current generators at a respective signal line (206). Each set (204) of programmable-resistance memory cells (203) includes a memory cell associated with each signal line that, in use, can be connected between the relevant signal line and a reference voltage so as to generate a voltage on the signal line. An adder module (207) is coupled to each of the signal lines to generate a voltage at an output node (210) based on the sum of the voltages on each of the signal lines.

Abstract: The present disclosure provides a method for converting numerical values into spikes. The method includes: generating an initial spike sequence according to input numerical values, where the initial spike sequence includes at least one data string, each of which is independently selected from one of a consecutive spike train or a consecutive non-spike train, the number of spikes in all of the at least one data string is equal to an expected value of the number of spikes in a target spike sequence to be generated, and the target spike sequence to be generated is a spike sequence of a spiking neural network within a time period; and randomly selecting and modifying data from the initial spike sequence, so as to form the target spike sequence. The present disclosure further provides an apparatus for converting numerical values into spikes, an electronic device, and a computer-readable storage medium.

Abstract: Methods and systems for performing a sequence of machine learning tasks. One system includes a sequence of deep neural networks (DNNs), including: a first DNN corresponding to a first machine learning task, wherein the first DNN comprises a first plurality of indexed layers, and each layer in the first plurality of indexed layers is configured to receive a respective layer input and process the layer input to generate a respective layer output; and one or more subsequent DNNs corresponding to one or more respective machine learning tasks, wherein each subsequent DNN comprises a respective plurality of indexed layers, and each layer in a respective plurality of indexed layers with index greater than one receives input from a preceding layer of the respective subsequent DNN, and one or more preceding layers of respective preceding DNNs, wherein a preceding layer is a layer whose index is one less than the current index.

Abstract: Provided are a classification model training method, system, electronic device, and storage medium. The method includes: determining sampling rates of first-class samples and second-class samples in a data set, and setting the samples with a sampling rate less than a preset value as target samples (S101); determining data distribution feature information of the target samples based on Euclidean distances between all the samples in the data set (S102); wherein the data distribution feature information is information describing the number of same-class samples in nearest neighbor samples, and the nearest neighbor samples are two samples at a Euclidean distance less than a preset distance; generating new samples corresponding to the target samples based on the data distribution feature information (S103); and training the classification model using the first-class samples, the second-class samples and the new samples (S104).

Abstract: A method includes deploying a neural network (NN) model on an electronic device. The NN model is generated by training a first NN architecture on a first dataset. A first function defines a first layer of the first NN architecture. The first function is constructed based on approximating a second function applied by a second layer of a second NN architecture. Retraining of the NN model is enabled on the electronic device using a second data set.

Abstract: A code review process utilizes a deep learning model trained on historical code reviews to automatically perform peer or code review of a source code file. The deep learning model is able to predict the code reviews relevant to a source code snippet by learning from historical code reviews. The deep learning model is trained on pairs of code snippets and code reviews that are relevant to each other and pairs of code snippets and code reviews that have no relation to each other. The deep learning model is data driven thereby not relying on pre-configured rules which makes the model adaptable to different review environments.

Abstract: A Thing Machine is provided having a processor, non-transitory memory, non-transitory computer readable media, and performable machine code P(TM). The P(TM) is comprised of a first set of performable machine code actions, having one or more performable machine code P(TM(i)) action, wherein each performable machine code P(TM(i)) action is configured as an implementation of an algorithmic procedure of a model, wherein a first P(TM(i)) provides an action of self-configuring a first vocabulary of Things in said non-transitory memory of the Thing Machine, said Things representative of Things that said processor can perform as actions, and the set of Things an action can act upon, and wherein at least one P(TM(i)) machine code action is performed to configure a second vocabulary of Things in the non-transitory memory of the Thing Machine representative of a core vocabulary through which an application can be provided.

Abstract: Systems and methods by a telematics server are provided. The method includes receiving, over a network, training data including model input data and a known output label corresponding to the model input data from a first device, training a centralized machine-learning model using the training data, determining, by the centralized machine-learning model, an output label prediction certainty based on the model input data, determining an increase in the output label prediction certainty over a prior predicted output label certainty of the centralized machine-learning model, and sending, over the network, a machine-learning model update to a second device in response to determining that the increase in the output label prediction certainty is greater than an output label prediction increase threshold.