Abstract: Provided is a spike neural network circuit including a synapse configured to generate an operation signal based on an input spike signal and a weight, and a neuron configured to generate an output spike signal using a comparator configured to compare a voltage of a membrane signal generated based on the operation signal with a voltage of a threshold signal, wherein the comparator includes a bias circuit configured to conditionally supply a bias current of the comparator depending on the membrane signal.

Abstract: Techniques are disclosed for identifying multimodal subevents within an event having spatially-related and temporally-related features. In one example, a system receives a Spatio-Temporal Graph (STG) comprising (1) a plurality of nodes, each node having a feature descriptor that describes a feature present in the event, (2) a plurality of spatial edges, each spatial edge describing a spatial relationship between two of the plurality of nodes, and (3) a plurality of temporal edges, each temporal edge describing a temporal relationship between two of the plurality of nodes. Furthermore, the STG comprises at least one of: (1) variable-length descriptors for the feature descriptors or (2) temporal edges that span multiple time steps for the event. A machine learning system processes the STG to identify the multimodal subevents for the event. In some examples, the machine learning system comprises stacked Spatio-Temporal Graph Convolutional Networks (STGCNs), each comprising a plurality of STGCN layers.

Abstract: A method and a system are for providing a medical data structure for a patient. The system includes a plurality of data sources, each data source to provide medical data of the patient; a computing device to implement an artificial neural network structure a plurality of encoding modules, each being realized as an artificial neural network configured and trained to generate, from the medical data from the corresponding data source, a corresponding encoded output matrix; a weighting gate module for each of the encoding modules; a concatenation module configured to concatenate weighted output matrices of the weighting gates to a concatenated output matrix; and an aggregation module realized as an artificial neural network configured and trained to receive the concatenated output matrix and to generate therefrom the medical data structure for the patient, the artificial neural network structure being trained as a whole using a cost function.

Abstract: Systems, methods and apparatus of intelligent bandwidth allocation to different types of operations to access storage media in a data storage device. For example, a data storage device of a vehicle includes: storage media components; a controller configured to store data into and retrieve data from the storage media components according to commands received in the data storage device; and an artificial neural network configured to receive, as input and as a function of time, operating parameters indicative a data access pattern, and generate, based on the input, a prediction to determine an optimized bandwidth allocation scheme for controlling access by different types of operations in the data storage device to the storage media components. The controller is configured to schedule the operations of the different types to access the one or more storage media components according to the optimized bandwidth allocation scheme.

Abstract: Aspects of the present disclosure relate to automatically generating a user manual using a technique that includes training a first model with a first set of training data. The technique further includes generating, by the first model, a set of operations and a set of windows, where the set of operations and the set of windows are functions of the program. The technique further includes, generating a plurality of tasks, where a first task comprises a first operation being performed on a first window. The technique further includes determining an order of the plurality of tasks and calculating a level score for the first operation of the first window. The technique further includes assembling the user manual having the plurality of tasks in the determined order.

Abstract: Because digital assistants tend to have different areas of expertise and/or different abilities to fulfill a given request, it is sometimes difficult for a user to know which digital assistant is best able to fulfill a request. Representative embodiments disclose mechanisms to increase federate digital assistants so that a user's request can be funneled to the digital assistant best able to fulfill the user's request. A meta-assistant gathers information on skills provided by a set of digital assistants. The meta-assistant also gathers completion data for requests for different digital assistants and user satisfaction information. A user submits a request to the meta-assistant. The meta-assistant extracts user intent from the request and redirects the user's request to the digital assistant best able to fulfill the request. Embodiments can utilize trained machine learning models or scored algorithmic approaches to select the digital assistant.

Abstract: A learning-type signal separation method performed using a model formulation unit, which performs learning processing based on a training-use signal including a specific component, and a training-use signal not including the specific component, the training-use signals including a common characteristic. The learning-type signal separation method includes: generating learned data by causing the model formulation unit to perform learning processing based on the training-use signal and information indicating whether or not the specific component is included in the training-use signal, to generate a data series signal in which the specific component has been separated and removed from a data series of the training-use signal; acquiring an arbitrary signal including the common characteristic; and generating, based on the acquired arbitrary signal and the generated learned data, a data series signal in which the specific component has been separated and removed from a data series of the arbitrary signal.

Abstract: According to an embodiment, a neural network apparatus includes cores, routers, a tree path, and a short-cut path. The cores are provided according to leaves in a tree structure, each core serving as a circuit that performs calculation or processing for part of elements of the neural network. The routers are provided according to nodes other than the leaves in the tree structure. The tree path connects the cores and the routers such that data is transferred along the tree structure. The short-cut path connects part of the routers such that data is transferred on a route differing from the tree path. The routers transmit data output from each core to any of the cores serving as a transmission destination on one of routes in the tree path and the short-cut path such that the calculation or the processing is performed according to a structure of the neural network.

Abstract: In an example embodiment, a system is provided whereby a machine learning model is trained to predict a standardization for a given raw title. A neural network may be trained whose input is a raw title (such as a query string) and a list of candidate titles (either title identifications in a taxonomy, or English strings), which produces a probability that the raw title and each candidate belong to the same title. The model is able to standardize titles in any language included in the training data without first having to perform language identification or normalization of the title. Additionally, the model is able to benefit from the existence of “loan words” (words adopted from a foreign language with little or no modification) and relations between languages.

Abstract: Provided is a processor implemented method that includes performing training or an inference operation with a neural network by obtaining a parameter for the neural network in a floating-point format, applying a fractional length of a fixed-point format to the parameter in the floating-point format, performing an operation with an integer arithmetic logic unit (ALU) to determine whether to round off a fixed point based on a most significant bit among bit values to be discarded after a quantization process, and performing an operation of quantizing the parameter in the floating-point format to a parameter in the fixed-point format, based on a result of the operation with the ALU.

Abstract: Techniques are generally described for causal impact estimation using machine learning. A first machine learning model is trained using non-treatment variables during training. A second machine learning model uses learned weights from the first machine learning model for non-treatment variables and is trained on one or more treatment variables. The second machine learning model estimates outcomes based on the presence or absence of an event represented by the treatment variable. Selection bias is reduced by warm-starting the second machine learning model with non-treatment variable weights learned during training of the first machine learning model.

Abstract: A processor-implemented neural network method includes calculating individual update values for a weight assigned to a connection relationship between nodes included in a neural network; generating an accumulated update value by accumulating the individual update values in an accumulation buffer; and training the neural network by updating the weight using the accumulated update value in response to the accumulated update value being equal to or greater than a threshold value.

Abstract: Embodiments of the present invention provide a method and apparatus for accelerating distributed training of a deep neural network. The method comprises: based on parallel training, the training of deep neural network is designed as a distributed training mode. A deep neural network to be trained is divided into multiple sub-networks. A set of training samples is divided into multiple subsets of samples. The training of the deep neural network to be trained is performed with the multiple subsets of samples based on a distributed cluster architecture and a preset scheduling method. The multiple sub-networks are simultaneously trained so as to fulfill the distributed training of the deep neural network. The utilization of the distributed cluster architecture and the preset scheduling method may reduce, through data localization, the effect of network delay on the sub-networks under distributed training; adapt the training strategy in real time; and synchronize the sub-networks trained in parallel.

Abstract: Mechanisms are provided for synthesizing a computer implemented neural network. An initially trained neural network is received and modified by introducing a new hidden layer of neurons and new connections that connect the new hidden layer of neurons to an output layer and a previous layer of neurons previously directly connected to the output layer of neurons to generate a modified neural network. The modified neural network is trained through one or more epochs of machine learning to generate modified weight values for the new connections and the new connections are pruned based on the modified weight values to remove a subset of the new connections and leaving remaining connections in the modified neural network. A merge operation is performed on the remaining connections in the modified neural network to generate a custom convolution filter and modified neural network. The modified neural network is then retrained for deployment.

Abstract: Authenticity of Artificial Intelligence (AI) results may be verified by creating, for an AI system, from a plurality of original inputs to form a plurality of original inference results, a plurality of original signatures of representative elements of an internal state of the AI system constructed from each individual original inference result of the plurality of original inference results. During deployment of the AI system, a matching of a plurality of deployment time inference results with a plurality of deployment time signatures, to the plurality of original signatures and the plurality of original inference results, may be verified.

Abstract: A method for verifying trusted communication between an agent device and an application providing apparatus using a registry apparatus. The registry apparatus maintains a device registry comprising authentication information for uniquely authenticating at least one agent device. The method includes the steps of obtaining from the device registry the authentication information for the agent device identified by a device identifier specified in an the authentication request from the agent device, performing verification of the agent device using the authentication information obtained from the device registry, and if the verification is not successful, transmitting to at least one of the agent device and the application providing apparatus revocation information for denying the trusted communication between the agent device and the application providing apparatus.

Abstract: In some examples, a data backup system may comprise a removable data storage item, wherein a manufacturer of the removable data storage item creates and stores an encryption key on the removable data storage item before the removable data storage item is shipped to an end user; a tamper-evident packaging including the removable data storage item, wherein the removable data storage item comprises a decryption key stored on a memory device accessible by disturbing the tamper-evident packaging; and a data transfer device to receive the removable data storage item, read the encryption key from the removable data storage item, encrypt backup data using the encryption key, and store the encrypted backup data on the removable data storage item.

Abstract: Single sign on technology enables shared access to a protected service, such as an application, from a plurality of dynamically associated computing devices. After logging into the application from one of the computing device, a user may access the application from the other computing devices without re-authentication. A user may also log out from the application from any of the computing device. Unique machine identifications, such as device DNA, for identifying each of the computing devices are used in, for example, a method, apparatus (such as a login server) and computer program product. A single session may be shared across multiple computing devices. The same authentication token, such as a SAML token, may also be used for all of the computing devices having the same user session.

Abstract: A computer-implemented method for controlling the expression of a block of data from a sensitive data storage device, the method including the steps of receiving from a software application a request to transfer the block of data from the source sensitive data storage device for expression at a destination device, determining a data mask indicator for the block of data, applying a limited expression format based upon the data mask indicator, and expressing the block of data at the destination device in the limited expression format, such as to facilitate protecting or masking sensitive data. The method may further include allowing a user to request revelation of a masked portion of the block of data, recording in a memory log user activity relating to such revelation request(s) of the user, and providing regular reports and/or administrative alerts relating to such logged user activity.

Abstract: Various technologies described herein pertain to utilization of shared Wi-Fi. For instance, network access rights of a Wi-Fi network can be controlled by a mobile device of a point of contact for the Wi-Fi network. Moreover, utilization of a Wi-Fi network can be tracked and usage data indicative of historic utilization of the Wi-Fi network can be retained. Further, groups of users between whom Wi-Fi credentials are shared can be created.