Abstract: A method of predicting a format of a file includes calculating a quotient vector storing a list of values in slots. Each slot of the quotient vector and binary vector correspond to a character of a set of distinct characters based on an order. Each value in the quotient vector indicates a frequency with which the corresponding distinct character is found relative to the set of distinct characters found in a file. The method also includes calculating, based on comparing each value in the quotient vector to a threshold, a binary vector storing a list of values in slots. The method further includes predicting, based on the binary vector, a format of the file.

Abstract: A system and method for detecting synthetic identities are provided that determine a synthetic identity score for a given user, the synthetic identity score indicating a likelihood that the given user is using a synthetic identity to conduct activities. The synthetic identity score generated by the system and method disclosed herein can then be used to determine a risk associated with the given user and to inform what actions to take based on the associated risk that the given user may use the synthetic identity to perform a bad act.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for virtualizing external memory as local to a machine learning accelerator. One ambient computing system comprises: an ambient machine learning engine; a low-power CPU; and an SRAM that is shared among at least the ambient machine learning engine and the low-power CPU; wherein the ambient machine learning engine comprises virtual address logic to translate from virtual addresses generated by the ambient machine learning engine to physical addresses within the SRAM.

Abstract: An artificial intelligence system for training a classifier has a database of training data and a modeling system for building a classification model based on the training data. The database has a binary classification for each entity and binary tokens indicating whether or not one or more indicators about the entity are true. The classification model is based on a tempered indication of the tokens. The tempered indication is a ratio of a weighted sum of the tokens for each entity divided by a tempering factor for each of the entities. The tempering factor is a function of the unweighted sum of the tokens for each entity. Thus, the tempering factor will reduce the tempered indication when large numbers of low weight tokens are present so that the model does not over predict the probability of an entity being in the classification.

Abstract: A trained base model is distributed to a set of nodes. From a first node in the set of nodes, a first set of meta-metrics resulting from a transfer learning operation on the trained base model at the first node is collected. The transfer learning at the first node uses first local data available at the first node. The first node is clustered in a cluster with a second node from the set of nodes, in response to a meta-metric in the first set of meta-metrics being within a tolerance value of a corresponding meta-metric in a second set of meta-metrics collected from the second node. A normalized set of model parameters is constructed after an iteration of transfer learning or local learning at the first and second nodes. The normalized set of model parameters is distributed to the first node and the second node in the cluster.

Abstract: A computer-implemented method and computer system for supervised machine learning and classification comprises applying, by a dimensional visualization module incorporating a global objective function, an unsupervised dimension reduction algorithm to a dataset including a plurality of data points capable of visual representation to produce a dimensionally reduced dataset and parsing, by the dimensional visualization module, the dimensionally reduced dataset to a user interface module for visual display of the dimensionally reduced dataset.

Abstract: Embodiments are directed to managing operations. If Operations events are provided, event clusters may be associated with one or more Operations events, such that the Operations events may be associated with the event clusters based on characteristics of the Operations events. Metrics including resolution metrics, root cause analysis, notes, and other remediation information may be associated with the event clusters. Then a modeling engine may be employed to train models based on the Operations events, the event clusters, and the resolution metrics, such that the trained model may be trained to correlate and predict the resolution metrics from real-time Operations events. If real-time Operations events may be provided, the trained models may be employed to predict the resolution metrics that are associated with the real-time Operations events. If model performance degrades beyond accuracy requirements, new observations may be added to the training set and the model re-trained.

Abstract: Systems and methods for machine-learning augmented application monitoring are disclosed. In one embodiment, in an information processing apparatus comprising at least one computer processor and a memory, a method for machine-learning augmented application monitoring may include: (1) receiving data from a plurality of data sources; (2) storing the data in a data store; (3) transforming the data; (4) extracting one or more feature and metric from the transformed data; (5) feeding the feature and metric into a machine learning model to identify at least one contributing metric; and (6) associating the contributing metric with at least one incident.

Abstract: Methods, computer-readable media, and devices are disclosed for improving an object model based upon measurements of physical properties of an object via an unmanned vehicle using adversarial examples. For example, a method may include a processing system capturing measurements of physical properties of an object via at least one unmanned vehicle, updating an object model for the object to include the measurements of the physical properties of the object, where the object model is associated with a feature space, and generating an example from the feature space, where the example comprises an adversarial example. The processing system may further apply the object model to the example to generate a prediction, capture additional measurements of the physical properties of the object via the at least one unmanned vehicle when the prediction fails to identify that the example is an adversarial example, and update the object model to include the additional measurements.

Abstract: A decision aid method for determining an action to be implemented by a given competitive entity in a competitive system comprises the competitive entity and at least one other adverse competitive entity, the competitive entity being able to implement an action from among a set of predefined actions, each action providing a different expected gain as a function of the actions implemented by the adverse competitive entities.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an output sequence from an input sequence. In one aspect, one of the systems includes an encoder neural network configured to receive the input sequence and generate encoded representations of the network inputs, the encoder neural network comprising a sequence of one or more encoder subnetworks, each encoder subnetwork configured to receive a respective encoder subnetwork input for each of the input positions and to generate a respective subnetwork output for each of the input positions, and each encoder subnetwork comprising: an encoder self-attention sub-layer that is configured to receive the subnetwork input for each of the input positions and, for each particular input position in the input order: apply an attention mechanism over the encoder subnetwork inputs using one or more queries derived from the encoder subnetwork input at the particular input position.

Abstract: Described is a system for predicting an occurrence of large-scale events using social media data. A collection of time series is acquired from social media data related to an event of interest. The collection of time series is partitioned into time intervals and semantic features are extracted from the time intervals as a set of semantic intervals. The semantic features are encoded into a multilayer network. Subgraphs of the multilayer network are transformed into a state transition network. A prediction of a future event of interest is generated by analyzing the encoded network using the state transition network. Using the analyzed encoded network, a device is controlled based on the prediction of the future event of interest.

Abstract: Example implementations described herein involve a system for maintenance predictions generated using a single deep learning architecture. The example implementations can involve managing a single deep learning architecture for three modes including a failure prediction mode, a remaining useful life (RUL) mode, and a unified mode. Each mode is associated with an objective function and a transformation function. The single deep learning architecture is applied to learn parameters for an objective function through execution of a transformation function associated with a selected mode using historical data. The learned parameters of the single deep learning architecture can be applied with streaming data from with the equipment to generate a maintenance prediction for the equipment.

Abstract: A set of methods and corresponding systems that analyze, predict, or classify data. The improvements are the result of techniques that leverage the simultaneous evaluation of training attributes.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for identifying answers to questions using neural networks. One of the methods includes receiving an input text passage and an input question string; processing the input text passage using an encoder neural network to generate a respective encoded representation for each passage token in the input text passage; at each time step: processing a decoder input using a decoder neural network to update the internal state of the decoder neural network; and processing the respective encoded representations and a preceding output of the decoder neural network using a matching vector neural network to generate a matching vector for the time step; and generating an answer score that indicates how well the input text passage answers a question posed by the input question string.

Abstract: A neural network that may include multiple layers of neural cells; wherein a certain neural cell of a certain layer of neural cells may include a first plurality of one-bit inputs; an adder and leaky integrator unit; and an activation function circuit that has a one-bit output; wherein the first plurality of one-bit inputs are coupled to a first plurality of one-bit outputs of neural cells of a layer that precedes the certain layer; wherein the adder and leaky integration unit is configured to calculate a leaky integral of a weighted sum of a number of one-bit pulses that were received, during a time window, by the first plurality of one-bit inputs; and wherein the activation function circuit is configured to apply an activation function on the leaky integral to provide a one-bit output of the certain neural cell.

Abstract: A system identifies multiple causal anomalies in a power plant having multiple system components. The system includes a processor. The processor constructs an invariant network model having (i) nodes, each representing a respective system component and (ii) invariant links, each representing a stable component interaction. The processor constructs a broken network model having (i) the invariant network model nodes and (ii) broken links, each representing an unstable component interaction. The processor ranks causal anomalies in node clusters in the invariant network model to obtain anomaly score results. The processor generates, using a joint optimization clustering process applied to the models, (i) a model clustering structure and (ii) broken cluster scores.

Abstract: A computing system trains a reinforcement learning model comprising multiple different attention model components. The reinforcement learning model trains on training data of a first environment (e.g., a first traffic intersection). The reinforcement learning model trains by training a state attention computer model on the training data that weighs each of respective inputs of a respective state. The reinforcement learning model trains by training an action attention computer model that determines a probability of switching from a first action to a second action of the first set of the multiple candidate actions (e.g., changing traffic colors of traffic lights). Alternatively, or additionally, a computing system generates an indication of a selected outcome according to the reinforcement learning model and sends a selection output to the second environment (e.g., a second traffic intersection with more lanes than the first traffic intersection) to implement the selected action in the second environment.

Abstract: Disclosed are systems, techniques, and non-transitory storage media for predicting social media postings as being trusted news or a type of suspicious news. The systems, techniques, and non-transitory storage media are based on unique neural network architectures that learn from a combined representation including at least representations of social media posting content and a vector representation of communications among connected users.

Abstract: Artificial neural networks (ANNs) are a distributed computing model in which computation is accomplished with many simple processing units, called neurons, with data embodied by the connections between neurons, called synapses, and by the strength of these connections, the synaptic weights. An attractive implementation of ANNs uses the conductance of non-volatile memory (NVM) elements to record the synaptic weight, with the important multiply—accumulate step performed in place, at the data. In this application, the non-idealities in the response of the NVM such as nonlinearity, saturation, stochasticity and asymmetry in response to programming pulses lead to reduced network performance compared to an ideal network implementation.