Abstract: A single unified machine learning model (e.g., a neural network) is trained to perform both supervised event predictions and unsupervised time-varying clustering for a sequence of events (e.g., a sequence representing a user behavior) using sequences of events for multiple users using a combined loss function. The unified model can then be used for, given a sequence of events as input, predict a next event to occur after the last event in the sequence and generate a clustering result by performing a clustering operation on the sequence of events. As part of predicting the next event, the unified model is trained to predict an event type for the next event and a time of occurrence for the next event. In certain embodiments, the unified model is a neural network comprising a recurrent neural network (RNN) such as an Long Short Term Memory (LSTM) network.

Abstract: A method of wireless communication by a user equipment (UE) includes receiving, from a base station, a configuration to train a neural network for multiple different signal to noise ratios (SNRs) of a channel estimate for a wireless communication channel. The method also includes determining a current SNR of the channel estimate is above a first threshold value. The method further includes training the neural network based on the channel estimate, to obtain a first trained neural network. The method still further includes perturbing the channel estimate to obtain a perturbed channel estimate, and training the neural network based on the perturbed channel estimate, to obtain a second trained neural network. The method includes reporting, to the base station, parameters of the first trained neural network along with the channel estimate, and parameters of the second trained neural network.

Abstract: A system for monitoring an environment may include an input device for monitoring and capturing pattern-based states of a model of the environment. The system may also include a thalamobot embodied in at least a first processor in communication with the input device. The thalamobot may include at least one filter for monitoring captured data from the input device and for identifying at least one state change within the captured data. The system may also include at least one critic and/or at least one recognition system. The at least one filter forwards said at least one state change to the critic and/or recognition system. Novel schemes are introduced to allow processors to interconnect themselves into brain-like structures that contemplate both the environment and the model thereof, unifying disparate data into discoveries. The significance of such discoveries is recognized either through neural activation patterns or the topologies of interconnecting neural modules.

Abstract: A method and apparatus for identifying behavior of a target, and a radar system applied to an automated driving scenario include receiving a radar echo signal from a target, processing the radar echo signal to obtain time-frequency domain data, processing the time-frequency domain data to obtain signal attribute feature data representing a first feature of a radar echo signal attribute and linear prediction coefficient (LPC) feature data representing a second feature of the radar echo signal, inputting the signal attribute feature data and the LPC feature data into a behavior identification model, and outputting behavior information of the target.

Abstract: A system and method for controlling a non-tactile device including a receiving device configured to receive signals corresponding to a user's brain waves or movements, the brain waves or movements corresponding to a series of directional intentions, the intentions defining at least one line pattern, a processor configured to process the at least one line pattern, each of said at least one line patterns associated with an action of the device, and output a control signal to the non-tactile device related to the action.

Abstract: Depth sensors comprising a focal plane array with photosites (PSs) directed in different directions, each PS operable to detect light arriving from an instantaneous field of view (IFOV) of the PS, a readout-set of readout circuitries (ROCs), each ROC coupled to readout-group PSs by multiple switches and operable to output an electric signal indicative of an amount of light impinging on the readout-group PSs when the read-out group is connected to the respective ROC via at least one of the switches, a controller operable to change switching states of the switches, such that at different times different ROCs of the readout-set are coupled to the readout-group and are exposed to reflections from different distances, and a processor operable to obtain the electric signals from the readout-set indicative of detected levels of reflected light collected from the IFOVs of the readout-group and to determine depth information for an object.

Abstract: A system includes inputs, outputs, and nodes between the inputs and the outputs. The nodes include hidden nodes. Connections between the nodes are determined based on a gradient computable using symmetric solution submatrices.

Abstract: A system for data object compression and reduction includes to implement, in accordance with obtained optimization constraint data, an optimization procedure configured to determine an optimal set of adjustments to a set of data objects that maximizes reduction of both a data set aggregate magnitude and a data link composite magnitude for at least one pair of a plurality of data sources, the optimal set of adjustments including an offset of multiple data objects of data objects of same data object type and opposite polarity, and to store data indicative of the optimal set of adjustments to the set of data objects.

Abstract: Systems and components for use with neural networks. An execution block and a system architecture using that execution block are disclosed. The execution block uses a fully connected stack of layers and one output is a forecast for a time series while another output is a backcast that can be used to determine a residual from the input to the execution block. The execution block uses a waveform generator sub-unit whose parameters can be judiciously selected to thereby constrain the possible set of waveforms generated. By doing so, the execution block specializes its function. The system using the execution block has been shown to be better than the state of the art in providing solutions to the time series problem.

Abstract: A system for controlling heating, ventilation, or air conditioning (HVAC) equipment of a building includes one or more processing circuits configured to generate simulated building data using a simulation model of the building, pre-train a reinforcement learning (RL) model using the simulated building data, operate the HVAC equipment of the building using the RL model, and retrain the RL model using actual building data generated responsive to operating the HVAC equipment using the RL model.

Abstract: The current density distribution is determined in an electronic device including a first and a second electrode, and a layer of a 2-dimensional conductive material extending between the first and second electrode. The total current through the electrodes is measured, and then a first current measurement probe is placed at a plurality of positions near the interface between the 2D material and the first electrode. The probe is coupled to the same voltage as the first electrode. The same is done at the interface between the channel and the second electrode, by placing a second probe coupled to the same voltage as the second electrode. The boundary conditions are determined for the current, and assuming that the current density vector is normal to the interfaces, this yields the boundary conditions for the current density vector. Finally, the continuity equation is solved, taking into account the boundary conditions.

Abstract: In some examples, artificial intelligence based corpus enrichment for knowledge population and query response may include generating, based on annotated training documents, an entity and relation annotation model, identifying, based on application of the entity and relation annotation model to a document set that is to be annotated, entities and relations between the entities for each document of the document set to generate an annotated document set, and categorizing each annotated document into a plurality of categories. Artificial intelligence based corpus enrichment may include determining whether an identified category includes a specified number of annotated documents, and if not, additional annotated documents may be generated for the identified category that may represent a corpus.

Abstract: Systems and methods for implementing accelerated memory transfers in an integrated circuit includes configuring a region of memory of an on-chip data buffer based on a neural network computation graph, wherein configuring the region of memory includes: partitioning the region of memory of the on-chip data buffer to include a first distinct sub-region of memory and a second distinct sub-region of memory; initializing a plurality of distinct memory transfer operations from the off-chip main memory to the on-chip data buffer; executing a first set of memory transfer operations that includes writing a first set of computational components to the first distinct sub-region of memory, and while executing, using the integrated circuit, a leading computation based on the first set of computational components, executing a second set of memory transfer operations to the second distinct sub-region of memory for an impending computation.

Abstract: Deep Neural Networks (DNNs) for forecasting future data are provided. In one embodiment, a non-transitory computer-readable medium is configured to store computer logic having instructions that, when executed, cause one or more processing devices to receive, at each of a plurality of Deep Neural Network (DNN) forecasters, an input corresponding to a time-series dataset of a plurality of input time-series datasets. The instructions further cause the one or more processing devices to produce, from each of the plurality of DNN forecasters, a forecast output and provide the forecast output from each of the plurality of DNN forecasters to a DNN mixer for combining the forecast outputs to produce one or more output time-series datasets.

Abstract: Systems and methods for pre-processing data to facilitate efficient and accurate machine learning are provided. The data may include market data. The pre-processing may include partitioning the data into windows assigning categories to windows generate a series of vectors. The series of vectors then being input into a computer system that executes a machine learning algorithm to efficiently train a neural network used to identify structure or patterns therein.

Abstract: A method for training a generative adversarial network, in particular a Wasserstein generative adversarial network. The generative adversarial network includes a generator and a discriminator, the generator and the discriminator being artificial neuronal networks. The method includes training the discriminator. In the step of training the discriminator, a parameter of the discriminator is adapted as a function of a loss function, the loss function including a term that represents the violation of the Lipschitz condition as a function of a first input datum and a second input datum and as a function of a first output of the discriminator when processing the first input datum and a second output of the discriminator when processing the second input datum, the second input datum being created starting from the first input datum by applying the method of the virtual adversarial training.

Abstract: An embodiment of a semiconductor package apparatus may include technology to embed one or more trigger operations in one or more messages related to collective operations for a neural network, and issue the one or more messages related to the collective operations to a hardware-based message scheduler in a desired order of execution. Other embodiments are disclosed and claimed.

Abstract: A neural network learning mechanism has a device which perturbs analog neurons to measure an error which results from perturbations at different points within the neural network and modifies weights and biases to converge to a target.

Abstract: A method can include receiving channels of data from equipment responsive to operation of the equipment in an environment where the equipment and environment form a dynamic system; defining a particle filter that localizes a time window with respect to the channels of data; applying the particle filter at least in part by weighting particles of the particle filter using the channels of data, where each of the particles represents a corresponding time window; and selecting one of the particles according to its weight as being the time window of an operational state of the dynamic system.

Abstract: A non-volatile memory device includes arrays of non-volatile memory cells that are configured to the store weights for a recurrent neural network (RNN) inference engine with a gated recurrent unit (GRU) cell. A set three non-volatile memory arrays, such as formed of storage class memory, store a corresponding three sets of weights and are used to perform compute-in-memory inferencing. The hidden state of a previous iteration and an external input are applied to the weights of the first and the of second of the arrays, with the output of the first array used to generate an input to the third array, which also receives the external input. The hidden state of the current generation is generated from the outputs of the second and third arrays.

Abstract: Embodiments relate to a system, program product, and method for inducing creativity in an artificial neural network (ANN) having an encoder and decoder. Neurons are automatically selected and manipulated from one or more layers of the encoder. An encoded vector is sampled for an encoded image. Decoder neurons and a corresponding activation pattern are evaluated with respect to the encoded image. The decoder neurons that correspond to the activation pattern are selected, and an activation setting of the selected decoder neurons is changed. One or more novel data instances are automatically generated from an original latent space of the selectively changed decoder neurons.

Abstract: A receiver for use in a data channel on an integrated circuit device includes a non-linear equalizer having as inputs digitized samples of signals on the data channel, decision circuitry configured to determine from outputs of the non-linear equalizer a respective value of each of the signals, and adaptation circuitry configured to adapt parameters of the non-linear equalizer based on respective ones of the value. The non-linear equalizer may be a neural network equalizer, such as a multi-layer perceptron neural network equalizer, or a reduced complexity multi-layer perceptron neural network equalizer. A method for detecting data on a data channel on an integrated circuit device includes performing non-linear equalization of digitized samples of input signals on the data channel, determining from output signals of the non-linear equalization a respective value of each of the output signals, and adapting parameters of the non-linear equalization based on respective ones of the value.

Abstract: Examples described herein utilize multi-layer neural networks, such as multi-layer recurrent neural networks to estimate an error-reduced version of encoded data based on a retrieved version of encoded data (e.g., data encoded using one or more encoding techniques) from a memory. The neural networks and/or recurrent neural networks may have nonlinear mapping and distributed processing capabilities which may be advantageous in many systems employing a neural network or recurrent neural network to estimate an error-reduced version of encoded data for an error correction coding (ECC) decoder, e.g., to facilitate decoding of the error-reduced version of encoded data at the decoder. In this manner, neural networks or recurrent neural networks described herein may be used to improve or facilitate aspects of decoding at ECC decoders, e.g., by reducing errors present in encoded data due to storage or transmission.

Abstract: According to one embodiment, a method, computer system, and computer program product for navigating driverless vehicles is provided. The present invention may include ingesting data pertaining to the operation of the driverless vehicle, utilizing that data to predict tasks, which are driverless vehicle service tasks such as parking, maintenance, fueling, et cetera. The invention may further include determining the risk that a user may have need of the driverless vehicle, and scheduling the tasks to provide a balanced combination of convenience to the user, effective maintenance of the driverless vehicle, cost, and time. The method further includes navigating the driverless vehicle to accomplish the scheduled tasks.

Abstract: A device includes an input unit, a nonlinear converter, and an output unit. The nonlinear converter and the output unit are connected via a connection path having a delay mechanism that realizes a feedback loop giving a delay to a signal. The delay mechanism includes a conversion mechanism that generates a plurality of signals with different delay times using the signal output from the nonlinear converter, generates a new signal by superimposing the plurality of signals, and outputs the generated signal to the output unit.

Abstract: Techniques are provided for implementing a recurrent neuron (RN) using magneto-electric spin orbit (MESO) logic. An RN implementing the techniques according to an embodiment includes a first MESO device to apply a threshold function to an input signal provided at a magnetization port of the MESO device, and scale the result by a first weighting factor supplied at an input port of the MESO device to generate an RN output signal. The RN further includes a second MESO device to receive the RN output signal at a magnetization port of the second MESO device and generate a scaled previous RN state value. The scaled previous state value is a scaled and time delayed version of the RN output signal based on a second weighting factor. The RN input signal is a summation of the scaled previous state value of the RN with weighted synaptic input signals provided to the RN.

Abstract: Described herein are systems and methods for providing a natural language comprehension system that employs a two-stage process for machine comprehension of text. The first stage indicates words in one or more text passages that potentially answer a question. The first stage outputs a set of candidate answers for the question, along with a first probability of correctness for each candidate answer. The second stage forms one or more hypotheses by inserting each candidate answer into the question and determines whether a sematic relationship exists between each hypothesis and each sentence in the text. The second processing circuitry generates a second probability of correctness for each candidate answer and combines the first probability with the second probability to produce a score that is used to rank the candidate answers. The candidate answer with the highest score is selected as a predicted answer.

Abstract: An electronic circuit for enabling an efficient use of an oscillating neural network for feature recognition may be provided. The electronic circuit comprises a network of coupled voltage-controlled oscillators, wherein each of the voltage-controlled oscillators is adapted for receiving an edge input signal which is phase-shifted by a fraction of a period length of the voltage-controlled oscillators according to a signal strength of an analog input signal allotted to a respective one of the voltage-controlled oscillators, and an active output circuit. The active output circuit includes input terminals connected to selected ones of the voltage-controlled oscillators, an adder portion for adding input signals present at the input terminals, and a non-linear amplifier, an which input line is of the non-linear amplifier being connected to an output line of the adder portion, thereby an efficient use of an oscillating neural network.

Abstract: A computing network comprising a first layer comprising a first set of oscillators and a second layer comprising a second set of oscillators is provided. The computing network further comprises a plurality of adjustable coupling elements between the oscillators of the first set and a plurality of nonlinear elements. The nonlinear elements are configured to couple output signals of the first layer to the second layer. The computing network further comprises an encoding unit configured to receive input signals, convert the input signals into phase-encoded output signals, and provide the phase-encoded output signals to the first layer.

Abstract: A computer architecture for an and-or neural network is disclosed. A computing machine accesses an input vector. The input vector comprises a numeric representation of an input to a neural network. The computing machine provides the input vector to the neural network comprising a plurality of ordered layers. The plurality of ordered layers are alternating AND-layers and OR-layers. Each of the plurality of ordered layers receives input from a preceding layer and/or provides output to a next layer. The computing machine generates an output of the neural network based on an output of a last one of the plurality of ordered layers in the neural network.

Abstract: Processors and methods for neural network processing are provided. A method in a processor including a matrix vector unit is provided. The method includes receiving vector data and actuation vector data corresponding to at least one layer of a neural network model for processing using the matrix vector unit, where each of digital values corresponding to the vector data and the actuation vector data is represented in a sign magnitude format. The method further includes converting each of the digital values corresponding to at least one of the vector data or the actuation vector data to corresponding analog values and multiplying the vector data and the actuation vector data in an analog domain and providing corresponding multiplication results in a digital domain.

Abstract: Using a recurrent neural network (RNN) that has been trained to a satisfactory level of performance, highly discriminative features can be extracted by running a sample through the RNN, and then extracting a final hidden state hh where i is the number of instructions of the sample. This resulting feature vector may then be concatenated with the other hand-engineered features, and a larger classifier may then be trained on hand-engineered as well as automatically determined features. Related apparatus, systems, techniques and articles are also described.

Abstract: A differential recurrent neural network (RNN) is described that handles dependencies that go arbitrarily far in time by allowing the network system to store states using recurrent loops without adversely affecting training. The differential RNN includes a state component for storing states, and a trainable transition and differential non-linearity component which includes a neural network. The trainable transition and differential non-linearity component takes as input, an output of the previous stored states from the state component along with an input vector, and produces positive and negative contribution vectors which are employed to produce a state contribution vector. The state contribution vector is input into the state component to create a set of current states. In one implementation, the current states are simply output.

Abstract: An information processing method for system identification includes: generating a fitting curve represented by a sum of exponential functions for each of a set of digital inputs and a set of digital outputs for a physical system that is represented by one or plural equations including m-order differential operators (m is an integer equal to or greater than 1); and calculating coefficients of the differential operators, which are included in first coefficients, so that a first coefficient of each exponential function included in an expression obtained by a product of the differential operators and the fitting curve for the set of the digital inputs is equal to a second coefficient of the same exponential function, which is included in the fitting curve for the set of the digital outputs.

Abstract: A computer implemented method and apparatus for mitigating face aging errors when performing facial recognition. The method comprises receiving an indication of a face that needs to be searched in an image set, where each image in the image set comprises a timestamp that identifies a creation date of the image, the creation date being in a continuum of successive time intervals; and identifying the indicated face in images taken in each time interval of a plurality of successive time intervals for the indicated face, wherein each face found in images taken in a previous successive time interval is used as a reference set for identifying the face in images taken in a next successive time interval.

Abstract: A neural network designing method forms a RNN (Recurrent Neural Network) circuit to include a plurality of oscillating RNN circuits configured to output natural oscillations, and an adding circuit configured to obtain a sum of outputs of the plurality of oscillating RNN circuits, and inputs discrete data to the plurality of oscillating RNN circuits in order to compute a fitting curve with respect to the discrete data output from the adding circuit.

Abstract: A neural system comprises multiple neurons interconnected via synapse devices. Each neuron integrates input signals arriving on its dendrite, generates a spike in response to the integrated input signals exceeding a threshold, and sends the spike to the interconnected neurons via its axon. The system further includes multiple noruens, each noruen is interconnected via the interconnect network with those neurons that the noruen's corresponding neuron sends its axon to. Each noruen integrates input spikes from connected spiking neurons and generates a spike in response to the integrated input spikes exceeding a threshold. There can be one noruen for every corresponding neuron. For a first neuron connected via its axon via a synapse to dendrite of a second neuron, a noruen corresponding to the second neuron is connected via its axon through the same synapse to dendrite of the noruen corresponding to the first neuron.

Abstract: A system and device for solving sparse algorithms using hardware solutions is described. The hardware solution can comprise one or more analog devices for providing fast, energy efficient solutions to small, medium, and large sparse approximation problems. The system can comprise sub-threshold current mode circuits on a Field Programmable Analog Array (FPAA) or on a custom analog chip. The system can comprise a plurality of floating gates for solving linear portions of a sparse signal. The system can also comprise one or more analog devices for solving non-linear portions of sparse signal.

Abstract: A neural network designing method forms a RNN (Recurrent Neural Network) circuit to include a plurality of oscillating RNN circuits configured to output natural oscillations, and an adding circuit configured to obtain a sum of outputs of the plurality of oscillating RNN circuits, and inputs discrete data to the plurality of oscillating RNN circuits in order to compute a fitting curve with respect to the discrete data output from the adding circuit.

Abstract: A method for approximation of optimal control for a nonlinear discrete time system in which the state variables are first obtained from a system model. Control sequences are then iteratively generated for the network to optimize control variables for the network and in which the value for each control variable is independent of the other control variables. Following optimization of the control variables, the control variables are then mapped onto a recurrent neural network utilizing conventional training methods.

Abstract: A control system (1) for a complex process, particularly for controlling a combustion process in a power plant, a waste incinerator plant, or a cement plant, has a controlled system (14) and at least one controller (36), wherein the control system (1) is divided hierarchically into various levels (10, 20, 30, 40). The first level (10) represents the complex, real process to be controlled and is implemented by the controlled system (14). The second level (20) represents an interface to the process and is implemented by a process control system. The third level (30) represents the control of the process and is implemented by the at least one active controller (36). The fourth level (40) represents a superordinate overview and is implemented by a principal controller (44).

Abstract: A method for digitally signing of electronic documents which are to be kept secure for a very long time, thereby taking into account future cryptographic developments which could render currently cryptographic key-lengths insufficient. A double signature is issued for each document. A first digital signature ensures the long term security, while a second digital signature ensures the involvement of an individual user. Thereby, the second digital signature is less computationally intensive in its generation than the first digital signature.

Abstract: Systems and methods for a scalable artificial neural network, wherein the architecture includes: an input layer; at least one hidden layer; an output layer; and a parallelization subsystem configured to provide a variable degree of parallelization to the artificial neural network by providing scalability to neurons and layers. In a particular case, the systems and methods may include a back-propagation subsystem that is configured to scalably adjust weights in the artificial neural network in accordance with the variable degree of parallelization. Systems and methods are also provided for selecting an appropriate degree of parallelization based on factors such as hardware resources and performance requirements.

Abstract: A method for computer-aided control of any technical system is provided. The method includes two steps, the learning of the dynamic with historical data based on a recurrent neural network and a subsequent learning of an optimal regulation by coupling the recurrent neural network to a further neural network. The recurrent neural network has a hidden layer comprising a first and a second hidden state at a respective time point. The first hidden state is coupled to the second hidden state using a matrix to be learned. This allows a bottleneck structure to be created, in that the dimension of the first hidden state is smaller than the dimension of the second hidden state or vice versa. The autonomous dynamic is taken into account during the learning of the network, thereby improving the approximation capacity of the network. The technical system includes a gas turbine.

Abstract: Machine-readable media, methods, apparatus and system for obtaining and processing image features are described. In some embodiments, groups of training features derived from regions of training images may be trained to obtain a plurality of classifiers, each classifier corresponding to each group of training features. The plurality of classifiers may be used to classify groups of validation features derived from regions of validation images to obtain a plurality of weights, wherein each weight corresponds to each region of the validation images and indicates how important the each region of the validation images is. Then, a weight may be discarded from the plurality of weights based upon a certain criterion.

Abstract: A processor architecture for a learning machine is presented which uses a massive array of processing elements having local, recurrent connections to form global associations between functions defined on manifolds. Associations between these functions provide the basis for learning cause-and-effect relationships involving vision, audition, tactile sensation and kinetic motion. Two arbitrary input signals hold each other in place in a manifold association processor and form the basis of short-term memory.

Abstract: A system and method for remotely connecting client computers to a communication network such as the Internet by way of a server system handling a plurality of client computers and having the capability of dynamically providing network connections to the client computers, separately billing usage time and tracking usage and preferably updating access software on the client computers.

Abstract: A processor architecture for a learning machine is presented which uses a massive array of processing elements having local, recurrent connections to form global associations between functions defined on manifolds. Associations between these functions provide the basis for learning cause-and-effect relationships involving vision, audition, tactile sensation and kinetic motion. Two arbitrary images hold each other in place in a manifold association processor and form the basis of short-term memory.

Abstract: A data processing device for processing time-sequence data includes a data extracting unit extracting time-sequence data for a predetermined time unit from time-sequence data; and a processing unit obtaining scores for nodes of an SOM configured from multiple nodes provided with a spatial array configuration, the scores showing applicability to time-sequence data for a predetermined time unit thereof. The node with the best score is determined to be the winning node which is the node most applicable. The processing unit obtains scores as to the time-sequence data for one predetermined time unit, regarding a distance-restricted node wherein distance from the winning node as to the time-sequence for a predetermined time unit immediately preceding the time-sequence data of one predetermined time unit is within a predetermined distance. The distance-restricted node with the best the score is determined to be the winning node.

Abstract: A learning system is provided, which includes network storage means for storing a network including a plurality of nodes, each of which holds a dynamics; and learning means for self-organizationally updating the dynamics of the network on the basis of measured time-series data.