Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating discrete latent representations of input data items. One of the methods includes receiving an input data item; providing the input data item as input to an encoder neural network to obtain an encoder output for the input data item; and generating a discrete latent representation of the input data item from the encoder output, comprising: for each of the latent variables, determining, from a set of latent embedding vectors in the memory, a latent embedding vector that is nearest to the encoded vector for the latent variable.

Abstract: Techniques are disclosed for managing artificial intelligence model partitions for execution in an information processing system with edge computing resources. For example, a method comprises the following steps. An intermediate representation of an artificial intelligence model is obtained. A computation graph is generated based on the intermediate representation. The computation graph is partitioned into a set of partitions. The method then schedules the set of partitions for respective execution on a set of computing devices in an edge computing environment, and causes deployment of the set of partitions respectively to the set of computing devices for execution in the edge computing environment.

Abstract: Disclosed systems include an electrocardiogram sensor and a processing device operatively coupled to the electrocardiogram sensor. The processing device receives electrocardiogram data from the electrocardiogram sensor and applies a machine learning model to the received electrocardiogram data. The machine learning model has been trained based on previous electrocardiogram data of a plurality of subjects. The electrocardiogram data of the plurality of subjects have one or more associated analyte measurements. The processing device may determine an indication of a level of the analyte based on the electrocardiogram data.

Abstract: A method for compressing a neural network model includes acquiring a to-be-compressed neural network model. A first bit width, a second bit width and a target thinning rate corresponding to the to-be-compressed neural network model are determined. A target value is obtained according to the first bit width, the second bit width and the target thinning rate. Then the to-be-compressed neural network model is compressed using the target value, the first bit width and the second bit width to obtain a compression result of the to-be-compressed neural network model.

Abstract: Techniques for evaluating an output of a machine learning model and using the evaluation to retrain the machine learning model are described. For example, a data set that is output from a layer of the machine learning model is reduced to a 2-D or 3-D representation that is suitable for viewing. A user views the reduced data set in a viewing environment such as virtual reality or augmented reality. The user makes changes using that viewing environment. The changes are then used to retrain the machine learning model.

Abstract: A convolutional neural network (CNN)-based signal processing includes receiving of an encrypted output from a first layer of a multi-layer CNN data. The received encrypted output is subsequently decrypted to form a decrypted input to a second layer of the multi-layer CNN data. A convolution of the decrypted input with a corresponding decrypted weight may generate a second layer output, which may be encrypted and used as an encrypted input to a third layer of the multi-layer CNN data.

Abstract: Computer systems and computer-implemented methods modify a machine learning network, such as a deep neural network, to introduce judgment to the network. A “combining” node is added to the network, to thereby generate a modified network, where activation of the combining node is based, at least in part, on output from a subject node of the network. The computer system then trains the modified network by, for each training data item in a set of training data, performing forward and back propagation computations through the modified network, where the backward propagation computation through the modified network comprises computing estimated partial derivatives of an error function of an objective for the network, except that the combining node selectively blocks back-propagation of estimated partial derivatives to the subject node, even though activation of the combining node is based on the activation of the subject node.

Abstract: The present disclosure provides a meteorological big data fusion method based on deep learning, including the following steps: constructing multi-source meteorological data samples; according to an original resolution of different climate variables, selecting a corresponding super-resolution multiple to obtain an optimized super-resolution module under the constraint of maximizing information retention efficiency; constructing a spatial-temporal attention module using a focused attention mechanism, and selecting a corresponding time stride according to periodic characteristics of different climate variables; constructing a meteorological data fusion model in combination with the optimized super-resolution model and the spatial-temporal attention module; taking a minimum resolution of climate variables as a loss function, and training the meteorological data fusion model with the multi-source meteorological data samples; and importing the acquired real-time meteorological data from multiple data sources into t

Abstract: A system including one or more computers and one or more storage devices storing instructions that when executed by the one or more computers cause the one or more computers to implement a memory and memory-based neural network is described. The memory is configured to store a respective memory vector at each of a plurality of memory locations in the memory. The memory-based neural network is configured to: at each of a plurality of time steps: receive an input; determine an update to the memory, wherein determining the update comprising applying an attention mechanism over the memory vectors in the memory and the received input; update the memory using the determined update to the memory; and generate an output for the current time step using the updated memory.

Abstract: Computer systems and computer-implemented methods modify a machine learning network, such as a deep neural network, to introduce judgment to the network. A “combining” node is added to the network, to thereby generate a modified network, where activation of the combining node is based, at least in part, on output from a subject node of the network. The computer system then trains the modified network by, for each training data item in a set of training data, performing forward and back propagation computations through the modified network, where the backward propagation computation through the modified network comprises computing estimated partial derivatives of an error function of an objective for the network, except that the combining node selectively blocks back-propagation of estimated partial derivatives to the subject node, even though activation of the combining node is based on the activation of the subject node.

Abstract: Techniques for user customized private label prediction are described. According to some embodiments, customers can train a classifier to detect new objects in image data. These new objects may not be included in a base model provided by a service provider system. The base model can be utilized to perform object detection and feature extraction from training images that are annotated by the customer to identify the new objects. Once trained, the new custom model can be used to identify the new objects in input images and label the images accordingly.

Abstract: Systems and methods of an integrated technology platform create a marketplace providing dashboards configured to allow brands and social media influencers to directly connect with each other. The system includes an integrated platform that enables an advertising party to find social media influencers who are most suited to the brands' contexts, market appeal, and demographic targets, build and manage relationships with the influencers, and identify fake influencers using machine learning models.

Abstract: A neural network method of parameter quantization obtains channel profile information for first parameter values of a floating-point type in each channel included in each of feature maps based on an input in a first dataset to a floating-point parameters pre-trained neural network, and determines a probability density function (PDF) type, for each channel, appropriate for the channel profile information based on a classification network receiving the channel profile information as a dataset. The neural network method of parameter quantization determines a fixed-point representation, based on the determined PDF type, for each channel, statistically covering a distribution range of the first parameter values, and generates a fixed-point quantized neural network based on the fixed-point representation determined for each channel.

Abstract: In federated learning problems, data is scattered across different servers and exchanging or pooling it is often impractical or prohibited. A Bayesian nonparametric framework is presented for federated learning with neural networks. Each data server is assumed to provide local neural network weights, which are modeled through our framework. An inference approach is presented that allows us to synthesize a more expressive global network without additional supervision, data pooling and with as few as a single communication round. The efficacy of the present invention on federated learning problems simulated from two popular image classification datasets is shown.

Abstract: Computer systems and computer-implemented methods train a neural network, by: (a) computing for each datum in a set of training data, activation values for nodes in the neural network and estimates of partial derivatives of an objective function for the neural network for the nodes in the neural network; (b) selecting a target node of the neural network and/or a target datum in the set of training data; (c) selecting a target-specific improvement model for the neural network, wherein the target-specific improvement model, when added to the neural network, improves performance of the neural network for the target node and/or the target datum, as the case may be; (d) training the target-specific improvement model; (e) merging the target-specific improvement model with the neural network to form an expanded neural network; and (f) training the expanded neural network.

Abstract: Classification of an input task data set by meta level continual learning includes analyzing first and second training data sets in a task space to generate first and second meta weights and a slow weight value, and comparing an input task data set to the slow weight to generate a fast weight. The first and second meta weights are parameterized with the fast weight value to update the slow weight value, whereby a value is associated with the input task data set, thereby classifying the input task data set by meta level continual learning.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for improving operational efficiency within a data center by modeling data center performance and predicting power usage efficiency. An example method receives a state input characterizing a current state of a data center. For each data center setting slate, the state input and the data center setting slate are processed through an ensemble of machine learning models. Each machine learning model is configured to receive and process the state input and the data center setting slate to generate an efficiency score that characterizes a predicted resource efficiency of the data center if the data center settings defined by the data center setting slate are adopted t. The method selects, based on the efficiency scores for the data center setting slates, new values for the data center settings.

Abstract: Techniques for generating counterfactuals in connection with machine learning models. The techniques include applying a trained machine learning model to an input to obtain a first outcome; determining whether the first outcome has a value in a set of one or more target values; when it is determined that the first outcome does not have a value in the set of one or more target values, generating a counterfactual input at least in part by applying a trained neural network model to the input to obtain a corresponding output, the corresponding output indicating changes to be made to one or more values of one or more attributes of the input to obtain the counterfactual input, and generating feedback based on the counterfactual input.

Abstract: Systems and methods for performing neural architecture search are provided. In one aspect, the system includes a processor configured to select a plurality of candidate neural networks within a search space, evaluate a performance of each of the plurality of candidate neural networks by: training each candidate neural network on a training dataset to perform the predetermined task and determining a ranking metric for each candidate neural network based on an objective function. The ranking metric includes a weight-related metric that is determined based on weights of a prediction layer of each respective candidate neural network before and after the respective candidate neural network is trained. The processor is configured to rank the plurality of candidate neural networks based on the determined ranking metrics.

Abstract: A method and system distinguish between anomalous members of a majority group and members of a target group. The system and method utilize a neural network architecture that attends to each level of a classification hierarchy. The system and method chain a semi-supervised autoencoder with a supervised classifier neural network. The autoencoder is trained in a semi-supervised manner with a machine learning process to identify user profile data that are typical of a majority class. The classifier neural network is trained in a supervised manner with a machine learning process to distinguish between user profile data that are anomalous members of the majority class and user profile data that are members of the target class.