Abstract: A prediction model (1) includes a first module (M1) that calculates, for each of a plurality of objects (xi) in a dataset (x), an index value (vi) corresponding to a combination of the object (xi) and attribute information (a) using a neural network, and a second module (M2) that calculates a prediction result (y) of an operation to be performed by a user by performing a predetermined process on a plurality of index values (v1, . . . , vN) obtained from the first module (M1) and corresponding to the respective plurality of objects (x1, . . . , xN).

Abstract: A temporal-aware or permutation-dependent Graph Neural Network (GNN) is disclosed. The example GNN is implemented by combining temporal-awareness with multi-layer neighborhood aggregation to further provide the GNN with inductive capabilities with respect to generating embeddings of a dynamic graph, all without creating multiple time snapshots of the graph. By using a temporal-aware message pass scheme involving a temporal-aware and permutation-dependent GNN, a set of temporal-aware local neighborhood aggregator functions may be effective trained and used for generating embeddings for unknow nodes and for providing more accurate embeddings for subsequent prediction tasks.

Abstract: An agent training method includes: obtaining environment information of a first agent and environment information of a second agent; generating first information based on the environment information of the first agent and the environment information of the second agent; and training the first agent by using the first information, so that the first agent outputs individual cognition information and neighborhood cognition information. The neighborhood cognition information of the first agent is consistent with neighborhood cognition information of the second agent.

Abstract: A neural network operation apparatus may include a receiver configured to receive input data to perform the neural network operation and a quantized Look Up Table (LUT) corresponding to a non-linear function comprised in the neural network operation, and a processor configured to perform scale-up on the input data based on a scale factor, to extract a quantized LUT parameter from the quantized LUT based on scaled-up input data, and to generate an operation result by performing a neural network operation based on the quantized LUT parameter.

Abstract: A method for approximating an activation function, the method including: receiving an input value of the activation function; determining that the input value is within a range, the range includes a set of non-uniform intervals; determining a selected interval from among the set of non-uniform intervals including the input value; retrieving, by a hardware accelerator, from a look-up table (LUT) associated with a type of the activation function, values of one or more quadratic interpolation parameters associated with the selected interval; performing a quadratic interpolation on the input value to approximate the input value using the values of the one or more quadratic interpolation parameters; and determining a first approximated output of the activation function based on a result of the quadratic interpolation performed on the input value.

Abstract: Systems and methods herein describe an image compression system. The image compression system generates a first generative adversarial network (GAN), identifies a threshold, based on the threshold, generates a second GAN by pruning channels of the first GAN, trains the second GAN using similarity-based knowledge distillation from the first GAN, and stores the trained second GAN.

Abstract: A reservoir includes a common input layer, first and second output layers that outputs a first and a second readout values based on an input, a first partial reservoir including the input layer and the first output layer, and a second partial reservoir having a size between the input layer and the second output layer larger than the size of the first partial reservoir, and the training processing including: first calculating a third output weight that reduces a difference between a first product sum value of a third readout value and a first output weight; and second calculating a fourth output weight that reduces a difference between a second product sum value of a fourth readout value and a second output weight and differential teaching data that is a difference between a third product sum value of the third readout value and the third output weight and the teaching data.

Abstract: Systems and methods for adaptive calibration of quantum computing systems are disclosed. A system can include one or more processors coupled to non-transitory memory and configured to obtain telemetry data from a quantum processor of a quantum computing system, generate a spatio-temporal graph data structure based on the telemetry data and operational parameters, and provide at least a portion of the graph as input to a graph neural network (GNN) to generate parameter ranges for operation. The system can select a first set of test parameters using a Bayesian optimization function, execute calibration experiments to generate calibration results, generate updated operational parameters based on the results, and control the quantum processor according to the updated parameters.

Abstract: Systems and methods herein describe an image compression system. The image compression system generates a first generative adversarial network (GAN), identifies a threshold, based on the threshold, generates a second GAN by pruning channels of the first GAN, trains the second GAN using similarity-based knowledge distillation from the first GAN, and stores the trained second GAN.

Abstract: The present disclosure relates to a method of inter-layer format conversion for a neural network, the neural network comprising at least two computation layers including a first layer to process first data in a first data format and a second layer to process second data in a second data format, the method comprising: extracting data statistics from data output by the first layer, said data statistics being representative of the data output by the first layer; determining one or more conversion parameters based on the extracted data statistics and the second data format; and generating the second data for the second layer by modifying said data output by the first layer using the one or more conversion parameters.

Abstract: A system for cognitive inferencing for large language models (LLMs), the system including: a integration layer communicatively connected to a static LLM, wherein the integration layer includes a processor configured to receive one or more user requests from a user, wherein the one or more user requests include a user interaction with the static LLM, generate one or more cognitive inference (CI) units from the one or more user requests, assign a confidence-falsifiability (CF) delta for each of the one or more CI units using a Socratic engine and append the one or more CI units and the CF delta to a cognitive inference (CI) log associated with the user, wherein the CI log serves as an inference engine for the static LLM, wherein the integration layer is configured to query the CI log upon a subsequent user interaction with the static LLM.

Abstract: An information management system is disclosed herein that can use artificial intelligence to identify situations in which a performance metric may not be satisfied. For example, a storage manager of the information management system can maintain data related to historical, current, and/or future execution of secondary copy operations by secondary storage computing device(s) in the information management system. Using some or all of this data, the storage manager can train an artificial intelligence model (e.g., a neural network) to classify whether a current or future secondary copy operation job is likely to succeed or fail. Similarly, the storage manager can use some or all of this data to train another artificial intelligence model (e.g., a machine learning model) to predict the length of time for a current or future secondary copy operation job to complete. The trained models can be used to predict whether a performance metric will be satisfied.

Abstract: Methods, systems, and computer programs are presented for predicting a response probability to a sent notification. One method includes an operation for training respective neural networks to obtain a first, second, and third models. The first model generates an embedding based on member information. The second and third model generate parameters for a distribution function. The first model is used to calculate a member embedding when accessing a notification for a member. Further, the method second model calculates a first parameter value, and the third model calculates a second parameter value based on the member embedding. Further, the method determines, a first probability that the member will visit the online service in response to the notification and a second probability that the member will visit without sending the notification. The method further includes determining to send the notification based on the first probability and the second probability.

Abstract: Methods and systems for implementing sedenion value neural network models are provided. A method includes obtaining, from a data store or a user device, values associated with a feature for which a value is to be determined. The method includes generating an input sedenion using the values and inputting the input sedenion to a sedenion value neural network (SVNN) model configured to generate an output sedenion using the input sedenion. The output sedenion may indicate a sedenion representation of the value to be determined. The method includes generating a sequence of real values using the output sedenion. The sequence may be or include the value. The method also includes outputting, to the data store or to the user device, the sequence of real values.

Abstract: In one embodiment, a device detects an object in an area based on sensor data generated by a plurality of sensors deployed to that area. The device uses a semantic reasoning engine to determine an appearance type of the object, based on the sensor data. The device uses the semantic reasoning engine to determine a behavioral type of the object, based on the sensor data. The device makes, using the semantic reasoning engine, a determination that the behavioral type of the object does not match the appearance type of the object. The device provides an indication of the determination for display.

Abstract: A system including a main neural network for performing one or more machine learning tasks on a network input to generate one or more network outputs. The main neural network includes a Mixture of Experts (MoE) subnetwork that includes a plurality of expert neural networks and a gating subsystem. The gating subsystem is configured to: apply a softmax function to a set of gating parameters having learned values to generate a respective softmax score for each of one or more of the plurality of expert neural networks; determine a respective weight for each of the one or more of the plurality of expert neural networks; select a proper subset of the plurality of expert neural networks; and combine the respective expert outputs generated by the one or more expert neural networks in the proper subset to generate one or more MoE outputs.

Abstract: This disclosure relates generally to system and method for determining explainability of machine predicted decisions. Typical explainable AI (XAI) solutions are limited by type of data processed, such as structured, semi-structured and unstructured text. In addition, due to limited automation of the process of explainability, typical systems are cumbersome and time-consuming. The system and method provide an end to end solution for automating the determination of explainability of machine predicted decisions. The XAI process output an absolute relevance score indicative of relevance of the features associated with the prediction which is indicative of percentage relevance/contribution of individual feature. The system further computes relative relevance score of the features by adding up all the features and calculating how much each individual feature is contributing to the total score. The relative relevance scores are utilized for determining explainability of decisions of the prediction.

Abstract: A hierarchical neural network for predicting out of stock products comprises an input layer that receives data from data sources that store disparate datasets having different levels of attribute detail pertaining to products for sale in stores of a retailer. A first level of neural networks processes the data from the data sources into respective learned intermediate vector representations. A second level comprises a concatenate layer that concatenates the learned intermediate vector representations from the second level into a combined vector representation. A third level comprises a feed forward network that receives the combined vector representation and outputs to the retailer an out of stock probability indicating which store and product combinations are likely to have out of stock products over a predetermined timeframe.

Abstract: A processing system including at least one processor may obtain description information of a first machine learning model, obtain a set of interpretation criteria for the first machine learning model, and generate, via a second machine learning model, an explanation text providing an interpretation of the first machine learning model in accordance with the set of interpretation criteria and the description information of the first machine learning model.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training neural networks with label differential privacy. One of the methods includes, for each training example: processing the network input in the training example using the neural network in accordance with the values of the network parameters as of the beginning of the training iteration to generate a network output, generating a private network output for the training example from the target output in the training example and the network output for the training example, and generating a modified training example that includes the network input in the training example and the private network output for the training example; and training the neural network on at least the modified training examples to update the values of the network parameters.