Abstract: A method includes sending a data selection parameter to a client computing device, wherein the client computing device stores a distributed instance of a machine learning model, and the machine learning model includes at least a first subset of parameters and a second subset of parameters. In response to a determination that a transmission criterion is satisfied, the client computing device is caused to report the first subset of parameters based on the data selection parameter. The method further comprises obtaining the parameters of the first subset of parameters of the distributed instance from the client computing device and updating the federated learning model based on the first subset of parameters of the distributed instance from the client computing device.

Abstract: Methods and systems are provided for a differential equations network. In one example, the differential equations network comprises one or more neuron within a single neural layer, where each of the neurons is configured to learn an activation function different or similar to an activation function learned by a different neuron within the same layer.

Abstract: Disclosed herein include systems, devices, and methods for flexible machine learning by traversing functionally invariant paths in weight space.

Abstract: Methods and systems are presented for configuring and training a machine learning model using transfer learning techniques that can transfer knowledge among multiple domains that do not share an identical feature set. Instead of using any feature set associated with a domain, a feature arrangement that combines all of the feature sets associated with the multiple domains in a particular organization for configuring and training the machine learning model. The feature arrangement includes a domain independent section and multiple domain-specific sections corresponding to the multiple domains. The domain independent section includes common features that are common across the multiple domains. Each of the domain-specific sections includes a feature set associated with the corresponding domain. The machine learning model that is configured in this manner can be trained to learn knowledge across the multiple domains and subsequently perform tasks for the multiple domains.

Abstract: A spiking neural network device according to an embodiment includes a synaptic element, a neuron circuit, a determinator, a synaptic depressor, and a synaptic potentiator. The synaptic element has a variable weight and outputs, in response to input of a first spike signal, a synaptic signal having intensity adjusted in accordance with the weight. The neuron circuit outputs a second spike signal in a case where the synaptic signal is inputted and a predetermined firing condition for the synaptic signal is satisfied. The determinator determines whether or not the weight is to be updated on a basis of an output frequency of the second spike signal by the neuron circuit. The synaptic depressor performs depression operation for depressing the weight in a case where it is determined that the weight is to be updated. The synaptic potentiator performs potentiating operation for potentiating the weight.

Abstract: One embodiment sets forth a technique for creating a generative model. The technique includes generating a trained generative model with a first component that converts data points in the training dataset into latent variable values, a second component that learns a distribution of the latent variable values, and a third component that converts the latent variable values into output distributions. The technique also includes training an energy-based model to learn an energy function based on values sampled from a first distribution associated with the training dataset and values sampled from a second distribution during operation of the trained generative model. The technique further includes creating a joint model that includes one or more portions of the trained generative model and the energy-based model, and that applies energy values from the energy-based model to samples from the second distribution to produce additional values used to generate a new data point.

Abstract: A method performed by a central server node is provided. The method includes: receiving local model weights and corresponding key from a local client node; and updating a model pool having a plurality of central models and corresponding keys associated with each of the central models. Updating the model pool is based on the local model weights, and one or more of the key corresponding to the local client node and the keys collectively corresponding to each of the central models. Updating the model pool comprises updating at least two of the plurality of central models contained in the model pool.

Abstract: A method and system for evaluating evaluation target by various characteristics using MOE are disclosed. According to one embodiment, an evaluation system may include Reader Modules for receiving evaluation target data and preprocessing it by characteristics of each of a plurality of artificial intelligence models for evaluation, an Evaluation Controller for controlling input of the preprocessed evaluation target data into the plurality of artificial intelligence models for evaluation, Multi-dimensional Evaluation Models for configuring each of the plurality of artificial intelligence models for evaluation into a plurality of instance models in multi-dimension by characteristics of the evaluation target data, and Federated Evaluation Models for integrating evaluation results for the preprocessed evaluation target data of each of the plurality of artificial intelligence models for evaluation and generating overall evaluation result.

Abstract: A neural processing device and method for pruning thereof are provided. The neural processing device includes a processing unit configured to perform calculations, an L0 memory configured to store input and output data of the processing unit, wherein the input and output data include a two-dimensional weight matrix and a weight manipulator configured to receive the two-dimensional weight matrix and partition it into preset sizes to thereby generate partitioned matrices, to generate a pruning matrix by pruning the partitioned matrix, and to transmit the pruning matrix to the processing unit.

Abstract: A training technique trains a neural network having sparsely-activated sub-networks. It does so by processing plural batches of training data in two respective passes of the neural network, yielding first prediction information and second prediction information. For each batch, the technique randomly assigns different sub-networks in the first and second passes of the neural network to process the batch. Over the course of training, the technique attempts to minimize loss information, which describes the difference between the first prediction information and ground-truth information, and the difference between the second prediction information and the ground-truth information. Simultaneously, the technique attempts to minimize divergence information, which describes the divergence of the first prediction information from the second prediction information (and vice versa).

Abstract: A method of using a computing device to improve an answer generated by a natural language question and answer system includes receiving, by a computing device, multiple questions in a natural language question and answer system. The computing device further generates multiple answers to the multiple questions. The computing device still further constructs a new training set with the generated multiple answers, where each answer is compared with a corresponding question of the multiple questions. The computing device additionally augments the new training set with one or more tokens delimiting a span of one or more of the generated multiple answers. The computing device further trains a new natural language question and answer system with the augmented new training set.

Abstract: Technological advancements are disclosed that utilize inertial sensor data for multiple classes to select a combination of filters to extract information though features to train a machine learning core decision tree. A determination is made whether the data for a class includes a frequency peak or dominating frequency that contains significant information about the class. In response to the data for the class including a frequency peak, a peak-based frequency range is determined. An entropy value is calculated for multiple frequency ranges in the data for the class. An entropy-based frequency range is selected from the multiple frequency ranges having a minimum entropy value. A frequency of interest is selected from the peak-based frequency range and the entropy-based frequency range for the class. A combination of filters is selected for each frequency of interest for each class and a decision tree is trained based on selected filter combination.

Abstract: The present invention relates to a method and a system for the segmentation of white matter hyperintensities (WMHs) present in magnetic resonance brain images, comprising: providing an array of trained convolutional neural networks (CNNs) with a magnetic resonance brain image; determining, for each of the CNNs and for each voxel, the probability that the given voxel corresponds to a pathological hyperintensity; calculating the average of all the probabilities determined for each voxel; comparing the averaged probabilities for each voxel with a threshold; generating an image mask with the voxels that exceed the threshold.

Abstract: A method for training a deep learning model may include: acquiring model description information and configuration information of a deep learning model; segmenting the model description information into at least two sections based on segmentation point variable in the configuration information, and loading the model description information to a corresponding resource to run; inputting a batch of training samples into a resource corresponding to a first section of model description information, then starting training and using obtained context information as an input of a resource corresponding to a subsequent section of model description information; and so on until an operation result of a resource corresponding to a final section of model description information is obtained; if a training completion condition is met, outputting a trained deep learning model; and otherwise, keeping on acquiring a subsequent batch of training samples and performing the above training steps until the condition is met.

Abstract: A variational auto-encoder model is trained to generate probabilities of action categories and probabilities of inter-arrival times of next action from a sequence of past actions by generating a concatenated representation of each action and associated time, encoding the concatenated representations, determining a conditional prior distribution for a next action, determining a conditional posterior distribution for the current action, sampling a latent variable from the conditional prior distribution, generating a probability distribution over a current action category, and generating a probability distribution over inter-arrival times for the current action category.

Abstract: Device and method for training an artificial neural network, including providing a neural network layer for an equivariant feature mapping having a plurality of output channels, grouping channels of the output channels into a number of distinct groups, wherein the output channels of each individual distinct group are organized into an individual grid defining a spatial location of each of the output channels of the individual distinct group in the grid for the individual distinct group, providing for each of the output channels of each individual distinct group, a distinct normalization function which is defined depending on the spatial location of the output channel in the grid in that this output channel is organized and depending on tunable hyperparameters for the normalization function, determining an output of the artificial neural network depending on a result of each of the distinct normalization functions, training the hyperparameters of the artificial neural network.

Abstract: A self-optimizing and self-programming computing system (SOSPCS) design framework that achieves both programmability and flexibility and exploits computing heterogeneity [e.g., CPUs, GPUs, and hardware accelerators (HWAs)] is provided. First, at compile time, a task pool consisting of hybrid tasks with different processing element (PE) affinities according to target applications is formed. Tasks preferred to be executed on GPUs or accelerators are detected from target applications by neural networks. Tasks suitable to run on CPUs are formed by community detection to minimize data movement overhead. Next, a distributed reinforcement learning-based approach is used at runtime to allow agents to map the tasks onto the network-on-chip-based heterogeneous PEs by learning an optimal policy based on Q values in the environment.

Abstract: One or more embodiments described herein facilitate identification and mitigation of cognitive bias in data-driven models. In one embodiment, a deep-learning system can comprise a memory that stores computer executable components; and a processor that executes the computer executable components stored in the memory. The computer executable components can comprise: an input component that receives data comprising primary task labels, secondary-identity attributes and a number of potential categories for one or more of the secondary-identity attributes; a machine-learning model that generates one or more predictions based on the received data; and a multi-objective learning component that trains the machine-learning model to mitigate bias from the one or more predictions.

Abstract: Systems and method are provided that are directed to tuning a hyperparameter associated with a small neural network model and transferring the hyperparameter to a large neural network model. At least one neural network model may be received along with a request for one or more tuned hyperparameters. Prior to scaling the large neural network, the large neural network is parameterized in accordance with a parameterizing scheme. The large neural network is then scaled and reduced in size such that a hyperparameter tuning process may be performed. A tuned hyperparameter may then be provided to a requestor such that the hyperparameter can be directly input into the large neural network. By tuning a hyper parameter using a small neural network, significant computation cycles and energy may be saved.

Abstract: In accordance with an embodiment, a method includes reducing a size of at least one initial parameter of each layer of an initial multilayer neural network to obtain for each layer a set of new parameters defining a new neural network, wherein each new parameter of the set of new parameters has its data represented in two portions comprising an integer portion and a fractional portion; implementing the new neural network using a test input data set applied only once to each layer; determining a distribution function or a density function resulting from the set of new parameters for each layer; and based on the determined distribution function or density function, adjusting a size of a memory area allocated to the fractional portion and a size of the memory area allocated to the integer portion of each new parameter associated with each layer.