Abstract: A method for accelerating score-based generative models (SGM) is provided, including setting a frequency mask (R) and a space mask (A) and a target sampling iteration number (T); sampling an initial sample (x0); conducting iteration comprising steps as follows: sampling a noise term; applying a preconditioned diffusion sampling (PDS) operator (M) to the noise term and thus generate a preconditioned noise term; calculating a drift term; applying the transpose of the PDS operator (MT) and then applying the PDS operator (M) to the drift term, and thus generate a preconditioned drift term; diffusing the sample of each iteration (xt); and outputting the result.

Abstract: A quantization parameter optimization method includes: determining a cost function in which a regularization term is added to an error function, the regularization term being a function of a quantization error that is an error between a weight parameter of a neural network and a quantization parameter that is a quantized weight parameter; updating the quantization parameter by use of the cost function; and determining, as an optimized quantization parameter of a quantization neural network, the quantization parameter with which a function value derived from the cost function satisfies a predetermined condition, the optimized quantization parameter being obtained as a result of repeating the updating, the quantization neural network being the neural network, the weight parameter of which has been quantized, wherein the function value derived from the regularization term and an inference accuracy of the quantization neural network are negatively correlated.

Abstract: Methods, devices, and systems for emulating a compute kernel with an ANN. The compute kernel is executed on a processor, and it is determined whether the compute kernel is a hotspot kernel. If the compute kernel is a hotspot kernel, the compute kernel is emulated with an ANN, and the ANN is substituted for the compute kernel.

Abstract: A storage device includes a memory device storing model information of a machine learning model; and a storage controller that controls an operation of the storage device using the machine learning model. The storage controller, upon receiving a get command for extracting the model information from the host device, reads the model information from the memory device in response to the get command and transmits the model information to the host device.

Abstract: A computer system is provided that is designed to handle multi-label classification. The computer system includes multiple processing instances that are arranged in a hierarchal manner and execute differently trained classification models. The classification task of one processing instance and the executed model therein may rely on the results of classification performed by another processing instance. Each of the models may be associated with a different threshold value that is used to binarize the probability output from the classification model.

Abstract: Techniques in advanced deep learning provide improvements in one or more of accuracy, performance, and energy efficiency. An array of processing elements performs flow based computations on wavelets of data. Each processing element has a compute element and a routing element. Each compute element has memory. Each router enables communication via wavelets with nearest neighbors in a 2D mesh. A compute element receives a wavelet. If a control specifier of the wavelet is a first value, then instructions are read from the memory of the compute element in accordance with an index specifier of the wavelet. If the control specifier is a second value, then instructions are read from the memory of the compute element in accordance with a virtual channel specifier of the wavelet. Then the compute element initiates execution of the instructions.

Abstract: In one embodiment, a method involves accessing training data, where the training data contains an ordered sequence of data associated with a plurality of entities, training one or more deep learning models to determine, from the ordered sequence of data, a first set of embeddings for each entity of the plurality of entities, where each entity has a plurality of entity attributes, determining, for each of the plurality of entity attributes, a corresponding initial embedding, training the one or more deep-learning models to refine the initial embeddings according to one or more criterion, generating one or more updated embeddings for each of the plurality of entities based on the refined initial embeddings of the plurality of entity attributes, and modifying the first set of embeddings based on the one or more updated embeddings.

Abstract: Batch normalization (BN) layer fusion and quantization method for model inference in artificial intelligence (AI) network engine are disclosed. A method for a neural network (NN) includes merging batch normalization (BN) layer parameters with NN layer parameters and computing merged BN layer and NN layer functions using the merged BN and NN layer parameters. A rectified linear unit (RELU) function can be merged with the BN and NN layer functions.

Abstract: Examples herein describe hardware architecture for processing and accelerating data passing through layers of a neural network. In one embodiment, a reconfigurable integrated circuit (IC) for use with a neural network includes a digital processing engine (DPE) array, each DPE having a plurality of neural network units (NNUs). Each DPE generates different output data based on the currently processing layer of the neural network, with the NNUs parallel processing different input data sets. The reconfigurable IC also includes a plurality of ping-pong buffers designed to alternate storing and processing data for the layers of the neural network.

Abstract: Embodiments described herein provide a system to configure distributed training of a neural network, the system comprising memory to store a library to facilitate data transmission during distributed training of the neural network; a network interface to enable transmission and receipt of configuration data associated with a set of worker nodes, the worker nodes configured to perform distributed training of the neural network; and a processor to execute instructions provided by the library. The instructions cause the processor to create one or more groups of the worker nodes, the one or more groups of worker nodes to be created based on a communication pattern for messages to be transmitted between the worker nodes during distributed training of the neural network. The processor can transparently adjust communication paths between worker nodes based on the communication pattern.

Abstract: Systems, apparatuses and methods may provide for technology that generates, by a full inference path of a neural network, a first detection result associated with one or more objects in a first video frame. The technology may also generate, by a partial inference path of the neural network, a second detection result based on the first detection result, wherein the second detection result corresponds to a second video frame that is subsequent to the first video frame.

Abstract: Systems and methods for performing improper input data detection are described. In one example, a system comprises: hardware circuits configured to receive input data and to perform computations of a neural network based on the input data to generate computation outputs; and an improper input detection circuit configured to: determine a relationship between the computation outputs of the hardware circuits and reference outputs; determine that the input data are improper based on the relationship; and perform an action based on determining that the input data are improper.

Abstract: The present invention provides a prediction method and system of high slope deformation. First, historical deformation data of each period of each part of a high slope is obtained as sample data; the sample data is divided into training samples and test samples; then a parameter group of a Support Vector Machine (SVM) model is optimized by using the training samples and a particle Swarm Optimization (PSO) algorithm to determine an optimal parameter group of the SVM model, to obtain a trained SVM model; whether the trained SVM model satisfies a condition is verified by using the test samples, and when the SVM model does not satisfy the condition, an optimal parameter group of the SVM model is re-determined; and finally the deformation of each area of the high slope is predicted by using the SVM model that satisfies the condition.

Abstract: A system and method for automatically generating recurrent neural networks for log anomaly detection uses a controller recurrent neural network that generates an output set of hyperparameters when an input set of controller parameters is applied to the controller recurrent neural network. The output set of hyperparameters is applied to a target recurrent neural network to produce a child recurrent neural network with an architecture that is defined by the output set of hyperparameters. The child recurrent neural network is then trained, and a log classification accuracy of the child recurrent neural network is computed. Using the log classification accuracy, at least one of the controller parameters used to generate the child recurrent neural network is adjusted to produce a different input set of controller parameters to be applied to the controller recurrent neural network so that a different child recurrent neural network for log anomaly detection can be generated.

Abstract: A data exchange method, a data exchange device, and a computing device for data exchange between a provider and a recipient for machine learning, the method including: (a) receiving a machine learning model from the provider (S1100); (b) respectively transforming output data samples into corresponding output eigenvectors by utilizing the machine learning model from the provider (S1200); (c) after transformation, combining the output eigenvectors with corresponding identifiers to form exchange samples (S1300). According to the data exchange method, original data is transformed into vector information which cannot be restored but can be applied to machine learning, for use in exchange, so as to, on one hand, enable efficient use of data for machine learning and, on the other hand, prevent unauthorized use, sale or disclosure of the original data.

Abstract: A computer-implemented method according to one embodiment includes receiving a single instance of training data, simplifying the single instance of training data to create a single instance of simplified training data, generating a plurality of training data variants, based on the single instance of simplified training data, and training a machine learning model, utilizing the plurality of training data variants.

Abstract: The present application describes a machine learning method for detecting tamper. The method includes a step of training a model using one or more values obtained from one or more different sensors on an integrated module. The one or more values act as training data with respect to one or more of light, acceleration, magnetic field, rotation, temperature, pressure, humidity, and audio. The method also includes a step of predicting, via the trained model, tampering of the of the integrated module. The present application also describes a system for detecting tamper.

Abstract: Systems, methods, and apparatuses for providing dynamic, prioritized spectrum utilization management. The system includes at least one monitoring sensor, at least one data analysis engine, at least one application, a semantic engine, a programmable rules and policy editor, a tip and cue server, and/or a control panel. The tip and cue server is operable utilize the environmental awareness from the data processed by the at least one data analysis engine in combination with additional information to create actionable data.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for selecting actions to be performed by an agent interacting with an environment. In one aspect, a system comprises a grid cell neural network and an action selection neural network. The grid cell network is configured to: receive an input comprising data characterizing a velocity of the agent; process the input to generate a grid cell representation; and process the grid cell representation to generate an estimate of a position of the agent in the environment; the action selection neural network is configured to: receive an input comprising a grid cell representation and an observation characterizing a state of the environment; and process the input to generate an action selection network output.

Abstract: Embodiments for deep learning for partial differential equation (PDE)-based models by a processor. A trained forecasting model and consistency constraints may be generated using a PDE-based model, a discretization of the PDE-based model, historical inputs the of the PDE-based model, and a representation of consistency constraints to generate a predictive output.