Abstract: A student model may be trained in two stages by using two teacher models, respectively. The first teacher model has been trained with a pretraining dataset. The second teacher model has been trained with a training dataset that is specific to a task to be performed by the student model. In the first stage, the student model may be generated based on a structure of the first teacher model. Internal parameters of the student model are adjusted through a pretraining process based on the first teacher model and the pretraining dataset. Weights of the student model may be pruned during the pretraining process. In the second stage, a sparsity mask is generated for the student model to lock the sparsity pattern generated from the first stage. Further, some of the internal parameters of the student model are modified based on the second teacher model and the training dataset.

Abstract: An apparatus and method for efficiently creating less computationally intensive nodes for a neural network. In various implementations, a computing system includes a memory that stores multiple input data values for training a neural network, and a processor. Rather than determine a bit width P of an integer accumulator of a node of the neural network based on bit widths of the input data values and corresponding weight values, the processor selects the bit width P during training. The processor adjusts the magnitudes of the weight values during iterative stages of training the node such that an L1 norm value of the weight values of the node does not exceed a corresponding weight magnitude limit.

Abstract: Generally discussed herein are devices, systems, and methods for machine learning (ML) modeling of a system that operates on a multivector object. A method includes receiving, by an ML model, the multivector object as an input that represents a state of the multivector system. The method includes operating, by the ML model and using a Clifford layer that includes neurons that implement a multivector kernel, on the multivector input to generate a multivector output that represents the state of the multivector system responsive to the multivector input.

Abstract: A neural network system and an operation method for a neural network system are provided. The neural network system includes at least one edge device and a server. Each edge device stores a neural network architecture. The neural network architecture includes at least one operator and a model identifier, and the at least one operator of the neural network architecture stored in the each edge device includes an operator identifier. The server is connected to the each edge device. The each edge device is configured to, upon being powered on, transmit the operator identifier of each operator to the server to request the server to return parameters for the each operator; receive the parameters of the each operator and combine the parameters of the each operator with the neural network architecture to obtain a neural network model; and execute a predetermined task based on the neural network model.

Abstract: A method for automated therapy discovery includes: accessing a corpus of scientific publications; compiling a population of semantic concepts from the corpus of scientific publications into a vector space model; deriving domains of concepts in the vector space model based on proximity to domain descriptors in the vector space model; deriving association scores and action characteristics between connected concepts, based on proximity and action descriptors in the vector space model; generating a semantic network; receiving a query for a target concept and a target domain at a research portal; isolating a set of edges between a target node and a subset of nodes; identifying subsets of concepts along the set of edges; generating hypotheses for directions and magnitudes of effects of subsets of concepts on the target concept based on association scores and action characteristics stored in connections along the set of edges; and returning hypotheses to the research portal.

Abstract: Disclosed are a method and an apparatus for real-time data monitoring based on machine learning, the method including: training a multi-layer predictor on actual values of historical indicator data, each layer of the multi-layer predictor including a plurality of predictors of different types; outputting predicted values of future indicator data by inputting a future time period for prediction into the trained multi-layer predictor; calculating alarm thresholds from the predicted values of the future indicator data and historical prediction errors; and triggering an alarm when an actual value of the future indicator data exceeds the corresponding alarm threshold. The accuracy of the alarm thresholds can be improved, and the alarm thresholds can be well adapted to the constantly changing indicator data. There is no need to manually configure a fixed alarm threshold, the accuracy of the alarm can be ensured, and the number of missed and false alarms can be reduced.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating representations of input sequences. One of the methods includes obtaining an input sequence, the input sequence comprising a plurality of inputs arranged according to an input order; processing the input sequence using a first long short term memory (LSTM) neural network to convert the input sequence into an alternative representation for the input sequence; and processing the alternative representation for the input sequence using a second LSTM neural network to generate a target sequence for the input sequence, the target sequence comprising a plurality of outputs arranged according to an output order.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training and deploying machine-learned communication over radio frequency (RF) channels. One method includes: determining an encoder and a decoder, at least one of which is configured to implement an encoding or decoding that is based on at least one of an encoder machine-learning network or a decoder machine-learning network that has been trained to encode or decode information over a communication channel; determining first information; using the encoder to process the first information and generate a first RF signal; transmitting, by at least one transmitter, the first RF signal through the communication channel; receiving, by at least one receiver, a second RF signal that represents the first RF signal altered by transmission through the communication channel; and using the decoder to process the second RF signal and generate second information as a reconstruction of the first information.

Abstract: According to one embodiment, a training device trains a first model. The first model estimates a period of a task from time-series data of an operation of a human. The device acquires first time-series data to which a label of the task is assigned. The device extracts a pattern from a period indicated by the label in the first time-series data. The pattern is used as a feature. The device generates timing data of an appearance timing of the pattern in the first time-series data. The device trains the first model by using the label, the first time-series data, and the timing data.

Abstract: Systems and methods for emulating a physical quantum system with a quantum computation. A model Hamiltonian that approximates a first quantization Hamiltonian of the physical quantum system is stored in memory. The physical system includes a plurality of particles. The first quantization Hamiltonian includes a plurality of first quantization energy operators, and the model Hamiltonian includes a plurality of energy terms corresponding to respective ones of the plurality of first quantization energy operators. Each energy term includes a respective energy operator, a respective energy register operator, and a respective inverse energy operator. The physical quantum system is emulated by performing a quantum computation on a plurality of qubits of the quantum computing system to emulate time evolution using the model Hamiltonian.

Abstract: Methods, systems, and apparatus for providing a ML model for inference, the ML model having been trained using a first set of training data to provide predictions associated with an adverse event, after training of the ML model, receiving data from one or more data sources, the data representative of characteristics relevant to predictions associated with the adverse event, providing a second set of training data, determining, by a trigger module, a trigger decision based on a set of signals at least partially determined from the second set of training data, the trigger decision indicating whether the ML model is to be one of updated and retrained based on the second set of training data, and selectively executing one of updating and retraining of the ML model using at least a portion of the second set of training data in response to the trigger decision.

Abstract: Disclosed is a novel neural network architecture and methods for generating neural network-based models from such architecture. A first version of the neural network, that is used for training purposes, includes one or more blocks in a first format that can then be replaced with corresponding blocks in a second format for execution. An executable model can thus be provided comprising a second version of the neural network including the one or more blocks in the second format. This then allows the training to be performed in a first, e.g. expanded format, but with a second, e.g. reduced, format model then provided for execution.

Abstract: A modular artificial neural sensing system includes a hierarchical network of neural sensing units including a neuromimetic sensor array of artificial sensory synapses and sensory neurons for receiving physicochemical sensed signals and for outputting sensor output signals. An artificial neural network processor is adapted for processing the sensor output signals and includes processor neurons interconnected by processor synapses forming first connections and second connections. The processor outputs processor output signals. A first sensor interface feeds processed or unprocessed sensed signals into the processor. A second sensor interface receives output predicted signals from other neural sensing units and feeds processed or unprocessed output predicted signals into the processor. A signal decoder decodes the processor output signals and outputs decoder output signals.

Abstract: A method for node cluster assignment in a graph includes initializing a plurality of wavefunctions, each one of the plurality of wavefunctions corresponding to nodes of the graph, constructing a plurality of quantum circuits, each corresponding to a graph Laplacian of the graph, evolving the plurality of wavefunctions at the plurality of quantum circuits, each one of the plurality of wavefunctions being evolved to a different time than other ones of the plurality of wavefunctions, measuring evolved states of the plurality of wavefunctions to generate a time-evolved wavefunction vector, and identifying a cluster assignment of a node of the graph based on the time-evolved wavefunction vector.

Abstract: Methods and systems for managing execution of inference models hosted by data processing systems are disclosed. To manage execution of inference models hosted by data processing systems, a system may include an inference model manager and any number of data processing systems. The inference model manager may communication system data for the communication system linking the data processing systems. The inference model manager may use the communication system data to determine whether the communication system meets inference generation requirements of the downstream consumer. If the communication system does not meet inference generation requirements of the downstream consumer, the inference model manager may obtain an inference generation plan to return to compliance with the inference generation requirements of the downstream consumer.

Abstract: Provided are systems and methods which more efficiency train embedding models through the use of a cache of item embeddings for candidate items over a number of training iterations. The cached item embeddings can be “stale” embeddings that were generated by a previous version of the model at a previous training iteration. Specifically, at each iteration, the (potentially stale) item embeddings included in the cache can be used when generating similarity scores that are the basis for sampling a number of items to use as negatives in the current training iteration. For example, a Gumbel-Max sampling approach can be used to sample negative items that will enable an approximation of a true gradient. New embeddings can be generated for the sampled negative items and can be used to train the model at the current iteration.

Abstract: The present invention relates to verification of damage to vehicles. More particularly, the present invention relates to a universal approach to automated generation of a damage estimate to a vehicle using images of the vehicle and verification of a manually-generated damage repair proposals using the automatically generated damage estimate. Aspects and/or embodiments seek to provide a computer-implemented method of generating one or more repair estimates from one or more photos of a damaged vehicle and comparing the generated estimate(s) to one or more input repair estimates to verify the one or more input repair estimates.

Abstract: Consistency metadata, including a parameter for a pseudo-random number source, are determined for training-and-evaluation iterations of a machine learning model. Using the metadata, a first training set comprising records of at least a first chunk is identified from a plurality of chunks of a data set. The first training set is used to train a machine learning model during a first training-and-evaluation iteration. A first test set comprising records of at least a second chunk is identified using the metadata, and is used to evaluate the model during the first training-and-evaluation iteration.

Abstract: A method, computer readable medium, and system are disclosed for training a neural network model. The method includes the step of selecting an input vector from a set of training data that includes input vectors and sparse target vectors, where each sparse target vector includes target data corresponding to a subset of samples within an output vector of the neural network model. The method also includes the steps of processing the input vector by the neural network model to produce output data for the samples within the output vector and adjusting parameter values of the neural network model to reduce differences between the output vector and the sparse target vector for the subset of the samples.

Abstract: One embodiment provides a graphics processor comprising an instruction cache to store an instruction and a compute block configured to perform multiply-accumulate operations in response to execution of the instruction. The compute block includes a scheduler to schedule a plurality of threads for execution of the instruction and multiply-accumulate circuitry configured to execute the instruction via the plurality of threads, wherein the multiply-accumulate circuitry includes a plurality of functional units configured to process, in parallel via the plurality of threads, a corresponding plurality of matrix elements to multiply a first matrix and a second matrix, and to multiply the first matrix and the second matrix includes to multiply data elements in a row of the first matrix by corresponding data elements in a column of the second matrix to generate a plurality of products.