Mitsubishi Electric Research Laboratories, Inc.

Mitsubishi Electric Corporation

Ye Wang

Toshiaki Koike-Akino

Shantanu Rane

Toshiaki Koike Akino

Stark C Draper

Prakash Ishwar

Philip Orlik

Kyeong Jin Kim

Keisuke Kojima

Yingheng Tang

Bing-Rong Lin

Kieran Parsons

Jianlin Guo

Pu Wang

Milutin Pajovic

Tim Marks

Takato Hirano

Nori Matsuda

Matthew Brand

Wei Sun

Mitsuhiro Hattori

Anoop Cherian

Rina Shimizu

Jianyuan Yu

Deniz Erdogmus

Ivan Sanz Gorrachategui

Andac Demir

Bryan Liu

Xiao Wang

Safa Medin

Qian Xie

Toshiaki Akino

Nisarg Jagdishbhai Raval

Ardhendu Tripathy

Xi Yu

Shuchin Aeron

Pierre Moulin

Hubert Dzieciol

A photonic device for splitting optical beams includes an input port configured to receive an input beam having an input power, a power splitter including perturbation segments arranged in a first region and a second region of a guide material having a first refractive index, each segment having a second refractive index, wherein the first region is configured to split the input beam into a first beam and a second beam, wherein and the second region is configured to separately guide the first and second beams, wherein the first refractive index is greater than the second refractive index, and output ports including first and second output ports connected the power splitter to respectively receive and transmit the first and second beams.

Generative model for inverse design of materials, devices, and structures

A system for automated construction of an artificial neural network architecture is provided. The system includes a set of interfaces and data links configured to receive and send signals, wherein the signals include datasets of training data, validation data and testing data, wherein the signals include a set of random number factors in multi-dimensional signals X, wherein part of the random number factors are associated with task labels Y to identify, and nuisance variations S. The system further includes a set of memory banks to store a set of reconfigurable deep neural network (DNN) blocks, hyperparameters, trainable variables, intermediate neuron signals, and temporary computation values including forward-pass signals and backward-pass gradients. The system further includes at least one processor, in connection with the interface and the memory banks, configured to submit the signals and the datasets into the reconfigurable DNN blocks, wherein the at least one processor is configured to execute a Bayesian graph exploration using the Bayes-Ball algorithm to reconfigure the DNN blocks such that redundant links are pruned to be compact by modifying the hyperparameters in the memory banks. The system realizes nuisance-robust variational Bayesian inference to be transferable to new datasets in semi-supervised settings.

Automated construction of neural network architecture with Bayesian graph exploration

A computer-implemented pre-processed time-delay autoencoder based anomaly detection method are provided for detecting anomalous states of machines arranged in a factory automation (FA) system or a manufacturing production line. The method includes acquiring source signals from the machines via an interface performing a data pre-processing process for the acquired source signals by normalizing value ranges of the acquired source signals and filtering undesired features from the acquired source signals performing a time-delayed data reform process for the pre-processed source signals based on a time-delay window to generate pre-processed time-delay data submitting pre-processed time-delay testing data to a pre-processed time-delayed autoencoder (Prep-TDAE) neural network, wherein the pre-processed time-delay testing data are collected online while the machines are operated, wherein the Prep-TDAE neural network has been pre-trained by using the pre-processed time-delay training data detecting, if an anomaly state is encountered with respect to the machines, by computing anomaly scores of the pre-processed time-delay testing data, and determining, when the anomaly state is detected, anomaly occurrence time, duration and severity with respect to the anomaly state of each of the machines.

Anomaly detection and diagnosis in factory automation system using pre-processed time-delay neural network with loss function adaptation

A system for fusion of Wi-Fi measurements from multiple frequency bands to monitor indoor and outdoor space is provided. The system includes a multi-band wireless network comprising a set of radio devices to provide coverage in an environment, wherein the set of radio devices are configured to establish wireless communication or sensing links over multi-band wireless channels, wherein the multi-band wireless channels use a first radio band at a millimeter wavelength and a second radio band at a centimeter wavelength. The system further includes a computing processor communicatively coupled to the set of radio devices and a data storage, wherein the data storage has data comprising a parameterized model, modules and executable programs. The computing processor is configured to receive measurement data over the multi-band wireless channels to obtain a set of heterogeneous sensor data, network transferred data, or wireless channel attribute data, fuse at least two types of measurements from the first and the second radio bands at one or more steps in a parameterized model to generate an estimated environmental state in the environment.

Multi-band Wi-Fi fusion for WLAN sensing

An anomaly detector for detecting anomaly in input data comprises an auto-encoder trained to encode the input data and decode the encoded input data to reconstruct the input data. Further, the anomaly detector comprises a classifier trained to determine a reconstruction loss indicative of a difference between the accepted input data and the reconstructed input data, where the reconstruction loss includes a weighted combination of a plurality of loss functions evaluating reconstruction losses of a plurality of parts of the reconstructed input data, different types of loss functions, or both. The classifier is further configured to detect an anomaly in the reconstructed input data when the reconstruction loss is above a threshold.

Apparatus and method for anomaly detection

Present disclosure discloses an image processing system and method for manipulating two-dimensional (2D) images of three-dimensional (3D) objects of a predetermined class (e.g., human faces). A 2D input image of a 3D object of the predetermined class is manipulated by manipulating physical properties of the 3D object, such as a 3D shape of the 3D input object, an albedo of the 3D input object, a pose of the 3D input object, and lighting illuminating the 3D input object. The physical properties are extracted from the 2D input image using a neural network that is trained to reconstruct the 2D input image. The 2D input image is reconstructed by disentangling the physical properties from pixels of the 2D input image using multiple subnetworks. The disentangled physical properties produced by the multiple subnetworks are combined into a 2D output image using a differentiable renderer.

System and method for manipulating two-dimensional (2D) images of three-dimensional (3D) objects

A system for flexible regularization and adaptable scaling of an artificial neural network is provided. The system includes a memory to store an artificial neural network and training data, a processor and interface to submit signals and training data into the neural network having a sequence of layers, each layer includes a set of neuron nodes, wherein a pair of nodes from neighboring layers are mutually connected with a plural of trainable parameters to pass the signals from the previous layer to next layer, a random number generator to modify the output signal of each neuron nodes for regularization in a stochastic manner following a multi-dimensional distribution across layer depth and node width directions of the neural network, wherein at least one layer has non-identical profile across neuron nodes, a training operator to update the neural network parameters by using the training data such that the output of neural network provides better values in a plural of objective functions; and an adaptive truncator to prune the output of neuron nodes at each layer in a compressed size of the neural network to reduce the computational complexity on the fly in downstream testing phase for any new incoming data.

Non-uniform regularization in artificial neural networks for adaptable scaling

Data communications and storage systems require error control techniques to be transferred successfully without failure. Polar coding has been used as a state-of-the-art forward error correction code for such an error control technique. However, the conventional decoding based on successive cancellation has a drawback in its poor performance and long latency to complete. Because the factor graph of polar codes has a lot of short cycles, a parallelizable belief propagation decoding also does not perform well. The method and system of the present invention provide a way to resolve the issues by introducing a protograph lifting expansion for a polar coding family so that highly parallelizable decoding is realized to achieve a high coding gain and high throughput without increasing the computational complexity and latency. The invention enables an iterative message passing to work properly by eliminating short cycles through a hill-climbing optimization of frozen bits allocation and permutation.

Protograph quasi-cyclic polar codes and related low-density generator matrix family

A battery diagnostic system is disclosed for an RUL estimation of a battery. The battery diagnostic system comprises a memory configured to store a neural network trained to estimate a remaining useful life (RUL) of a test battery from a predetermined set of features indicative of a battery cycle of the test battery and a capacity of the test battery; a charging system configured to charge and discharge the test battery to provide measurements of the battery cycle and the capacity of the test battery; a processor configured to extract the predetermined set of features from the measurements to submit the extracted set of features to the neural network and an output interface configured to output the estimated RUL of the test battery.

Battery diagnostic system for estimating remaining useful life (RUL) of a battery

A system is provided for training a device design network. The system includes an interface configured to input data of a device, a memory to store the device design network including an encoder, a decoder, and an adversarial block, and a processor. The processor is, in connection with the memory, configured to update the encoder and the decoder based on a first loss function to reduce the difference between the input data and output data of the decoder, wherein the encoder is constructed by at least one convolutional layer followed by at least one parallel fully connected layer to extract features of a layout of the device, and update the adversarial block to construct by maximizing a second loss function.

A communication system including a receiver to receive training data. An input interface to receive input data coupled to a hardware processor and a memory. The hardware processor is configured to initialize the privacy module using the training data. Generate a trained privacy module, by iteratively optimizing an objective function. Wherein for each iteration the objective function is computed by a combination of a distortion of the useful attributes in the transformed data and of a mutual information between the sensitive attributes and the transformed data. Such that the mutual information is estimated by the auxiliary module that maximizes a conditional likelihood of the sensitive attributes given the transformed data. Receive the input data via the input interface. Apply the trained privacy module on the input data to produce an application specific transformed data. A transmitter to transmit the application specific transformed data over a communication channel.

Data-driven privacy-preserving communication

A transmitter for transmitting an encoded codeword over a communication channel includes a source to accept source data, an irregular polar encoder operated by a processor to encode the source data with at least one polar code to produce the encoded codeword, a modulator to modulate the encoded codeword, and a front end to transmit the modulated and encoded codeword over the communication channel. The polar code is specified by a set of regular parameters including one or combination of parameters defining a number of data bits in the codeword, a parameter defining a data index set specifying locations of frozen bits in the encoded codeword, and a parameter defining a number of parity bits in the encoded codeword. The polar code is further specified by a set of irregular parameters including one or combination of parameters defining an irregularity of values of at least one regular parameter of the polar code, a parameter defining an irregularity of permutation of the encoded bits, a parameter defining an irregularity of polarization kernels in the polar code, and a parameter defining an irregularity in selection of de-activated exclusive-or operations on different stages of the polar encoding, and wherein the irregular polar encoder encodes the codeword using the regular and the irregular parameters of the polar code.

Irregular polar code encoding

Systems and methods for protecting a privacy of aggregated data from third parties, while preserving analytical usefulness of the aggregated data. A processor performs an offline training stage by, acquiring stored statistical contribution of sensitive devices that corresponds to the aggregated data, as a function of a state of a sensitive device at each time point within a period of time. Performing a real-time stage by determining a state of the sensitive device contributing to the aggregated data at each time point within the period of time. Selecting, from memory, based on the determined state of the sensitive device at each time point, a corresponding statistical contribution of the sensitive device to the aggregated data at each time point. Subtracting the statistical contribution selected for a time point from a value of the aggregated data at the corresponding time point, to produce modified aggregated data, and transmitting to third parties.

Method and systems using privacy-preserving analytics for aggregate data

A transmitter for transmitting an encoded codeword over a communication channel is described. The transmitter includes a source to accept source data, an irregular polar encoder operated by a processor to encode the source data with at least one polar code to produce an encoded codeword, a modulator to modulate the encoded codeword, and a front end to transmit the modulated and encoded codeword over the communication channel, wherein the polar code is specified by a set of regular parameters and a set of irregular parameters.

A receiver includes a polar decoder for decoding an encoded codeword transmitted over a communication channel. The receiver includes a front end to receive over a communication channel a codeword including a sequence of bits modified with noise of the communication channel and a soft decoder operated by a processor to produce a soft output of the decoding. The codeword is encoded by at least one polar encoder with a polar code. The processor is configured to estimate possible values of the bits of the received codeword using a successive cancelation list (SCL) decoding to produce a set of candidate codewords, determine a distance between each candidate codeword and a soft input to the soft decoder, and determine a likelihood of a value of a bit in the sequence of bits using a difference of distances of the candidate codewords closest to the received codeword and having opposite values at the position of the bit.

Growing community of inventors

Andover, MA, United States of America

Ye Wang