Abstract: Techniques for classifier generalization in a supervised learning process using input encoding are provided. In one aspect, a method for classification generalization includes: encoding original input features from at least one input sample {right arrow over (x)}S with a uniquely decodable code using an encoder E(?) to produce encoded input features E({right arrow over (x)}S), wherein the at least one input sample {right arrow over (x)}S comprises uncoded input features; feeding the uncoded input features and the encoded input features E({right arrow over (x)}S) to a base model to build an encoded model; and learning a classification function {tilde over (C)}E(?) using the encoded model, wherein the classification function {tilde over (C)}E(?) learned using the encoded model is more general than that learned using the uncoded input features alone.

Abstract: Embodiments of the present invention are directed to facilitating event forecasting. In accordance with aspects of the present disclosure, a set of events determined from raw machine data is obtained. The events are analyzed to identify leading indicators that indicate a future occurrence of a target event, wherein the leading indicators occur during a search period of time the precedes a warning period of time, thereby providing time for an action to be performed prior to an occurrence of a predicted target event. At least one of the leading indicators is used to predict a target event. An event notification is provided indicating the prediction of the target event.

Abstract: There is provided a computing apparatus that includes: a retaining unit configured to retain an approximation table that approximately represents an activation function of a neural network, the approximation table mapping between a plurality of discrete input samples of the activation function and output samples respectively corresponding to the plurality of input samples; and a computing unit configured to convert an input value of activation function computation to an output value using the approximation table retained by the retaining unit when the activation function is selected for the activation function computation. The plurality of input samples of the approximation table are set such that input samples more distant from a reference point in the domain of the activation function have a larger neighboring sample interval.

Abstract: A computing device identifies an anomaly among a plurality of observation vectors. An observation vector is projected using a predefined orthogonal complement matrix. The predefined orthogonal complement matrix is determined from a decomposition of a low-rank matrix. The low-rank matrix is computed using a robust principal component analysis algorithm. The projected observation vector is multiplied by a predefined demixing matrix to define a demixed observation vector. The predefined demixing matrix is computed using an independent component analysis algorithm and the predefined orthogonal complement matrix. A detection statistic value is computed from the defined, demixed observation vector. When the computed detection statistic value is greater than or equal to a predefined anomaly threshold value, an indicator is output that the observation vector is an anomaly.

Abstract: Embodiments include methods performed by a processor of a vehicle for allocating processing resources to concurrently-executing neural networks. The methods may include determining a priority of each of a plurality of neural networks executing on a vehicle processing system based on a contribution of each neural network to overall vehicle safety performance, and allocating computing resources to the plurality of neural networks based on the determined priority of each neural network. In some embodiments, the methods may dynamically adjust hyperparameters of one or more neural networks.

Abstract: Embodiments disclosed herein generally relate to a method and system for generating a container image. A computing system receives a request from a remote computer to provision a container comprising a machine learning model. The computing system generates a first API accessible by the remote computer. The computing system receives one or more parameters for the container via the API. The one or more parameters include a machine learning model type. The computing system retrieves from a library of a plurality of machine learning models a machine learning model corresponding to a type of model specified in the one or more parameters. The computing system generates a container image that includes the machine learning model. The computing system provisions a container based on the container image.

Abstract: Systems and methods for automatic recognition of entities related to cloud incidents are described. A method, implemented by at least one processor, for processing cloud incidents related information, including entity names and entity values associated with incidents having a potential to adversely impact products or services offered by a cloud service provider is provided. The method may include using at least one processor, processing the cloud incidents related information to convert at least words and symbols corresponding to a cloud incident into machine learning formatted data. The method may further include using a machine learning pipeline, processing at least a subset of the machine learning formatted data to recognize entity names and entity values associated with the cloud incident.

Abstract: Methods, systems, and apparatus for selecting actions to be performed by an agent interacting with an environment. One system includes a high-level controller neural network, low-level controller network, and subsystem. The high-level controller neural network receives an input observation and processes the input observation to generate a high-level output defining a control signal for the low-level controller. The low-level controller neural network receives a designated component of an input observation and processes the designated component and an input control signal to generate a low-level output that defines an action to be performed by the agent in response to the input observation.

Abstract: A method, a computer program product, and a system of adversarial semi-supervised one-shot training using a data stream. The method includes receiving a data stream based on an observation, wherein the data stream includes unlabeled data and labeled data. The method also includes training a prediction model with the labeled data using stochastic gradient descent based on a classification loss and an adversarial term and training a representation model with the labeled data and the unlabeled data based on a reconstruction loss and the adversarial term. The adversarial term is a cross-entropy between the middle layer output data from the models. The classification loss is a cross-entropy between the labeled data and an output from the prediction model. The method further includes updating a discriminator with middle layer output data from the prediction model and the representation model and based on a discrimination loss, and discarding the data stream.

Abstract: Disclosed are a method and apparatus for energy-aware deep neural network compression. A network pruning method for deep neural network compression includes measuring importance scores of a network unit by using an energy-based criterion with respect to a deep learning model, and performing network pruning of the deep learning model based on the importance scores.

Abstract: One or more computing devices, systems, and/or methods for generating time-preserving embeddings are provided. User trails of user activities performed by users are generated. Frequencies at which the activities were performed are identified. Indices are assigned to a set of activities identified from the activities as having frequencies above a threshold. Activity descriptions of the set of activities are mapped to the indices to generate a vocabulary. A model is trained using the user trails, timestamps of the activities, and the vocabulary to learn a set of time-preserving embeddings.

Abstract: According to an embodiments of the present disclosure, a method comprising: at an electronic device with one or more processors, obtaining a data set; identifying, based on the data set, a first data point set on a first embedding space, wherein each data point included in the first data point set corresponds to each data included in the data set; identifying a modified first data point set on the first embedding space based on the first data point set by adjusting a property associated with a distribution of the first data point set, wherein the modified first data point set includes at least one modified data point which is not included in the first data point set; and providing a Modified Image of Data (MIOD) by representing the modified first data point set on an imaging space may be provided.

Abstract: A method of modelling data, comprising: training an objective function of a linear classifier, based on a set of labeled data, to derive a set of classifier weights; defining a posterior probability distribution on the set of classifier weights of the linear classifier; approximating a marginalized loss function for an autoencoder as a Bregman divergence, based on the posterior probability distribution on the set of classifier weights learned from the linear classifier; and classifying unlabeled data using the autoencoder according to the marginalized loss function.

Abstract: Hardware implementations of, and methods for processing, a convolution layer of a DNN that comprise a plurality of convolution engines wherein the input data and weights are provided to the convolution engines in an order that allows input data and weights read from memory to be used in at least two filter-window calculations performed either by the same convolution engine in successive cycles or by different convolution engines in the same cycle. For example, in some hardware implementations of a convolution layer the convolution engines are configured to process the same weights but different input data each cycle, but the input data for each convolution engine remains the same for at least two cycles so that the convolution engines use the same input data in at least two consecutive cycles.

Abstract: Disclosed is a low-power and compact neuron circuit implementing a ReLU activation function including a first-layer synaptic array, a neuron transistor, a resistor, and a second-layer synaptic array. The neuron transistor is a MOS transistor having a threshold voltage-adjustable property, a gate electrode of the neuron transistor is connected to each voltage output end of the first-layer synaptic array, and a drain electrode of the neuron transistor is connected to each voltage input end of the second-layer synaptic array. Thus, it is possible to satisfy the decision computation and output of different synaptic array output values by adjusting the magnitude of the threshold voltage of the transistor. The neuron circuit requires only one transistor in cooperative connection with the first-layer synaptic array and the second-layer synaptic array to implement the ReLU activation function; therefore, a significant improvement is achieved in terms of energy efficiency, delay reduction, and space utilization.

Abstract: In one aspect, the invention comprises a system and method for control of a transaction state system utilizing a distributed ledger. First, the system and method includes an application plane layer adapted to receive instructions regarding operation of the transaction state system. Preferably, the application plane layer is coupled to the application plane layer interface. Second, a control plane layer is provided, the control plane layer including an adaptive control unit, such as a cognitive computing unit, artificial intelligence unit or machine-learning unit. Third, a data plane layer includes an input interface to receive data input from one or more data sources and to provide output coupled to a decentralized distributed ledger, the data plane layer is coupled to the control plane layer. Optionally, the system and method serve to implement a smart contract on a decentralized distributed ledger.

Abstract: In training a deep neural network using reduced precision, gradient computation operates on larger values without affecting the rest of the training procedure. One technique trains the deep neural network to develop loss, scales the loss, computes gradients at a reduced precision, and reduces the magnitude of the computed gradients to compensate for scaling of the loss. In one example non-limiting arrangement, the training forward pass scales a loss value by some factor S and the weight update reduces the weight gradient contribution by 1/S. Several techniques can be used for selecting scaling factor S and adjusting the weight update.

Abstract: A processor-implemented neural network method includes: obtaining a first weight kernel of a weight model and pruning information of the first weight kernel; determining, based on the pruning information, a processing range of an input feature map for each weight element vector of the first weight kernel; performing a convolution operation between the input feature map and the first weight kernel based on the determined processing range; and generating an output feature map of a neural network layer based on an operation result of the convolution operation.

Abstract: Implementations of the present disclosure include receiving a goal, providing a problem-specific knowledge graph that is responsive to at least a portion of the goal, determining a set of events from the problem-specific knowledge graph, processing data representative of events in the set of events through a first machine learning (ML) model to provide a set of event scores, each event score in the set of event scores being associated with a respective event in the set of events, determining a sub-set of events based on the set of event scores, for each event in the sub-set of events, determining at least one action by processing a sequence of actions through a second ML model, and outputting the sub-set of events and a set of actions for execution of at least one action in the set of actions.

Abstract: The present disclosure provides a composite machine learning system for a transaction labeling service. A transaction labeling service receives at least one descriptive string describing a transaction associated with a user. The service identifies a preliminary grouping from a generalized scheme. The service extracts a set of N-grams from the descriptive string and converts the N-grams and the preliminary grouping into a set of features. A machine learning model determines a label from a labeling scheme for the transaction based on the features. User input related to the label includes an accuracy indicator and a reliability indicator. If the reliability indicator satisfies a reliability condition, a set of training data for the machine learning model is updated based on the descriptive string and the label. The machine learning model is then trained against the updated set of training data.