Abstract: A system and method to reduce weight storage bits for a deep-learning network includes a quantizing module and a cluster-number reduction module. The quantizing module quantizes neural weights of each quantization layer of the deep-learning network. The cluster-number reduction module reduces the predetermined number of clusters for a layer having a clustering error that is a minimum of the clustering errors of the plurality of quantization layers. The quantizing module requantizes the layer based on the reduced predetermined number of clusters for the layer and the cluster-number reduction module further determines another layer having a clustering error that is a minimum of the clustering errors of the plurality of quantized layers and reduces the predetermined number of clusters for the another layer until a recognition performance of the deep-learning network has been reduced by a predetermined threshold.

Abstract: Provided is a state determination apparatus that appropriately performs pattern classification processing and/or pattern determination processing even when a map generated by the SOM technique includes discontinuous image regions. In the state determination apparatus, the matching processing unit obtains adaptability data indicating a correlation degree between template data indicating a state and the SOM output data. The state determination unit obtains a state evaluation value based on an activity value obtained by the activity value obtaining unit and the adaptability value. The time series estimation unit determines a state of an input data based on the state evaluation value and state transition probability between states. This allows for appropriately performing pattern classification processing and/or pattern determination processing even when a map generated by the SOM technique includes discontinuous image regions.

Abstract: A method and system are provided to calculate a future behavioral data and identify a relative causal impact of external factors affecting the data. Behavioral data and data for one or more external factors are harvested for a first time period. New behavioral data is harvested for a second time period. New data for the second time period is harvested. Based on a second training algorithm, a forecast time series value of a future behavioral data for a third time period that is after the second time period is calculated. A relative causal impact between each external factor and the predicted time series value of the behavioral data, for the third time period, is identified.

Abstract: A method for employee biometric tracking is provided. The method comprises providing to a user a plurality of wearable devices capable of being connected to the user, establishing a wireless connection between the plurality of wearable devices and a mobile device, collecting by the plurality of wearable devices a plurality of biometric data from the user, receiving by an application stored on the mobile device the plurality of biometric data, inputting into a predictive engine biometric data selected from the plurality of biometric data, determining by the predictive engine in response to the biometric data whether the user is at, or soon will be at, an alert level, creating an alert signal, and displaying the alert signal to the user.

Abstract: A system and method to reduce weight storage bits for a deep-learning network includes a quantizing module and a cluster-number reduction module. The quantizing module quantizes neural weights of each quantization layer of the deep-learning network. The cluster-number reduction module reduces the predetermined number of clusters for a layer having a clustering error that is a minimum of the clustering errors of the plurality of quantization layers. The quantizing module requantizes the layer based on the reduced predetermined number of clusters for the layer and the cluster-number reduction module further determines another layer having a clustering error that is a minimum of the clustering errors of the plurality of quantized layers and reduces the predetermined number of clusters for the another layer until a recognition performance of the deep-learning network has been reduced by a predetermined threshold.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for increasing the security of neural network by discretizing neural network inputs. One of the methods includes receiving a network input for a neural network; processing the network input using a discretization layer, wherein the discretization layer is configured to generate a discretized network input comprising a respective discretized vector for each of the numeric values in the network input; and processing the discretized network input using the plurality of additional neural network layers to generate a network output for the network input.

Abstract: A technique for executing a containerized stateful application that is deployed on a stateless computing platform is disclosed. The technique involves deploying a containerized stateful application on a stateless computing platform and executing the stateful application on the stateless computing platform. The technique also involves during execution of the stateful application, evaluating, in an application virtualization layer, events that are generated during execution of the stateful application to identify events that may trigger a change in state of the stateful application and during execution of the stateful application, updating a set of storage objects in response to the evaluations, and during execution of the stateful application, comparing events that are generated by the stateful application to the set of storage objects and redirecting a storage object that corresponds to an event to a persistent data store if the storage object matches a storage object in the set of storage objects.

Abstract: Various embodiments of the present invention relate generally to systems and processes for transforming a style of video data. In one embodiment, a neural network is used to interpolate native video data received from a camera system on a mobile device in real-time. The interpolation converts the live native video data into a particular style. For example, the style can be associated with a particular artist or a particular theme. The stylized video data can viewed on a display of the mobile device in a manner similar to which native live video data is output to the display. Thus, the stylized video data, which is viewed on the display, is consistent with a current position and orientation of the camera system on the display.

Abstract: Methods and apparatuses are described for generation of a data model for identifying and classifying training needs of individuals. A computer data store stores unstructured text. A server computing device generates a vector for search queries in the unstructured text, and generates a training course classification data model that comprises a multi-layered neural network. The server computing device executes the training course classification model using the vectors as input to generate a training course recommendation output vector. The server computing device updates the training course classification data model based upon a rating value for a training course.

Abstract: A method and system are provided. The method includes obtaining, by a hardware processor, candidate data representing a plurality of candidates. The method further includes calculating, by the hardware processor, for each of the candidates, a temporal next state of a Recurrent Neural Network (RNN) by inputting a corresponding one of the candidates to the RNN at a current state. The method also includes merging, by the hardware processor, the temporal next state for each of the candidates to obtain the temporal next state of the RNN.

Abstract: A machine learning hardware accelerator architecture and associated techniques are disclosed. The architecture features multiple memory banks of very wide SRAM that may be concurrently accessed by a large number of parallel operational units. Each operational unit supports an instruction set specific to machine learning, including optimizations for performing tensor operations and convolutions. Optimized addressing, an optimized shift reader and variations on a multicast network that permutes and copies data and associates with an operational unit that support those operations are also disclosed.

Abstract: A method of automating a fraud classification process includes generating or updating fraud classification rules, at least by training a machine learning program using fraud classifications of a plurality of financial accounts and financial transaction data associated with those accounts. The method also includes retrieving first financial transaction data associated with a first financial account, and selecting, by applying the fraud classification rules to the first financial transaction data, a first fraud classification. The first fraud classification may be selected from among a plurality of predetermined fraud classifications. The method also includes causing an indication of the first fraud classification to be displayed to one or more people via one or more respective computing device user interfaces, the indication further specifying at least the first financial account.

Abstract: A motion processing analysis method, system, and computer program product include selecting a first spatial graticule size, determining that a physical entity is associated with a first spatial graticule during a first timeframe and that the physical entity is associated with a second spatial graticule during a second timeframe, as a result of determining that the physical entity is associated with the first spatial graticule during the first timeframe and that the physical entity is associated with the second spatial graticule during the second timeframe, adding to a sum, determining that the sum is beyond an acceptable range, and as a result of determining that the sum beyond the acceptable range, selecting a second spatial graticule size.

Abstract: Methods, systems, and apparatuses for adapting a predictive model for a manufacturing process. One method includes receiving, with an electronic processor, a plurality of data points for a plurality of manufactured parts and the predictive model. The predictive model outputs a label for a manufactured part provided by the manufacturing process indicating whether the manufactured part is accepted or rejected. The method also includes estimating, with the electronic processor, a drift for each of the plurality of data points and generating, with the electronic processor, an adapted version of the predictive model based on the predictive model and the drift for each of the plurality of data points. In addition, the method includes outputting, with the electronic processor, a label for each of the plurality of manufactured parts using the adapted version of the predictive model.

Abstract: The present disclosure includes methods and systems for generating captions for digital images. In particular, the disclosed systems and methods can train an image encoder neural network and a sentence decoder neural network to generate a caption from an input digital image. For instance, in one or more embodiments, the disclosed systems and methods train an image encoder neural network (e.g., a character-level convolutional neural network) utilizing a semantic similarity constraint, training images, and training captions. Moreover, the disclosed systems and methods can train a sentence decoder neural network (e.g., a character-level recurrent neural network) utilizing training sentences and an adversarial classifier.

Abstract: A convolutional neural network is provided comprising artificial neurons arranged in layers, each comprising output matrices. An output matrix comprises output neurons and is connected to an input matrix, comprising input neurons, by synapses associated with a convolution matrix comprising weight coefficients associated with the output neurons of an output matrix. Each synapse consists of a set of memristive devices storing a weight coefficient of the convolution matrix. In response to a change of the output value of an input neuron, the neural network dynamically associates each set of memristive devices with an output neuron connected to the input neuron. The neural network comprises accumulator(s) for each output neuron; to accumulate the values of the weight coefficients stored in the sets of memristive devices dynamically associated with the output neuron, the output value of the output neuron being determined from the value accumulated in the accumulator(s).

Abstract: Technologies for optimization of machine learning training include a computing device to train a machine learning network with a training algorithm that is configured with configuration parameters. The computing device may perform many training instances in parallel. The computing device captures a time series of partial accuracy values from the training. Each partial accuracy value is indicative of machine learning network accuracy at an associated training iteration. The computing device inputs the configuration parameters to a feed-forward neural network to generate a representation and inputs the representation to a recurrent neural network. The computing device trains the feed-forward neural network and the recurrent neural network against the partial accuracy values. The computing device optimizes the feed-forward neural network and the recurrent neural network to determine optimized configuration parameters.

Abstract: A machine learning device acquires from a numerical controller information relating to machining when the machining is performed, and further acquires an actual delay time due to servo control and due to machine movement which are caused in the machining when the machining is performed. Then, the device performs supervised learning using the acquired machining-related information as input data, and using the acquired actual delay time due to servo control and due to machine movement as supervised data, and constructs a learning model, thereby predicting the machine delay time caused in a machine with high precision.

Abstract: Traffic condition forecasting techniques are provided that use matrix compression and deep neural networks. An illustrative method comprises obtaining a compressed origination-destination matrix indicating a cost to travel between pairs of a plurality of nodes, wherein the compressed origination-destination matrix is compressed using a locality-aware compression technique that maintains only non-empty data; obtaining a trained deep neural network trained using the compressed origination-destination matrix and past observations of traffic conditions at various times; and applying traffic conditions between two nodes in the compressed origination-destination matrix at a time, t, to the trained deep neural network to obtain predicted traffic conditions between the two nodes at a time, t+?. A tensor can be generated indicating an evolution of traffic conditions over a time span using a stacked Origination-Destination matrix comprising a plurality of past observations representing the time span.

Abstract: System and techniques for identifying novel information are described herein. A classified experience may be obtained. The classified experience may include a set of attributes. Memory counts of members of the set of attributes for a user may be obtained. A novelty score for the classified experience may be computed by comparing the set of attributes to the memory counts. The classified experience may be presented to the user when the novelty score meets a qualification criterion.