Abstract: An apparatus in one embodiment comprises a processing platform that includes a plurality of processing devices each comprising a processor coupled to a memory. The processing platform is configured to implement at least a portion of at least a first cloud-based system. The processing platform comprises a modelling language extension module configured to implement artificial intelligence-based decision points into a process flow and compile context attributes associated with the artificial intelligence-based decision points based on data from artificial intelligence systems.

Abstract: Reinforcement learning agents for resource allocation for iterative workloads, such as training Deep Neural Networks, are configured. One method comprises obtaining a specification of an iterative workload comprising multiple states and a set of available actions for each state, and a domain model of the iterative workload relating allocated resources with service metrics; adjusting weights of a reinforcement learning agent by performing iteration steps for each simulated iteration of the iterative workload and using variables from the simulated iteration to refine the reinforcement learning agent; and determining a dynamic resource allocation policy for the iterative workload.

Abstract: Implementations described and claimed herein provide systems and methods for developing a reservoir. In one implementation, a reservoir model is received. The reservoir model includes a static model and a dynamic model. The static model includes one or more clusters of a three-dimensional volume of the reservoir and an uncertainty quantification generated using a neural network. The dynamic model includes pressure values and fluid saturation values propagated across the three-dimensional volume through a nodal connectivity of neighboring clusters. A set of input features is generated from the static model and the dynamic model. The set of input features is related to a drilling attractiveness of a target region of the reservoir using a set of rules executed by a fuzzy inference engine. A quantification of the drilling attractiveness is generated. A recommendation for drilling in the reservoir is output based on the quantification of the drilling attractiveness.

Abstract: Anomalies in a data set may be difficult to detect when individual items are not gross outliers from a population average. Disclosed is an anomaly detector that includes neural networks such as an auto-encoder and a discriminator. The auto-encoder and the discriminator may be trained on a training set that does not include anomalies. During training, an auto-encoder generates an internal representation from the training set, and reconstructs the training set from the internal representation. The training continues until data loss in the reconstructed training set is below a configurable threshold. The discriminator may be trained until the internal representation is constrained to a multivariable unit normal. Once trained, the auto-encoder and discriminator identify anomalies in the evaluation set. The identified anomalies in an evaluation set may be linked to transaction, security breach or population trends, but broadly, disclosed techniques can be used to identify anomalies in any suitable population.

Abstract: A method is provided for commodity management. The method generates, using a Dynamic Boltzmann Machine (DyBM), a future mean prediction and a future standard deviation prediction of a financial time-series dataset for a commodity. The method measures, using Hellinger Distance (HD), an accuracy of the future mean prediction and the future standard deviation prediction. The method combines the future mean prediction and the future standard deviation prediction with the Hellinger Distance to determine a DyBM trustworthy prediction time period in which predictions by the DyBM, including the future mean prediction and the future standard deviation prediction, are deemed trustworthy. The method selectively performs an action relating to an ownership of the commodity based on at least one of the future mean prediction and the future standard deviation prediction, responsive to the future mean prediction and the future standard deviation prediction being generated during the DyBM trustworthy prediction time period.

Abstract: Systems for dynamically performing pattern recognition and data reconciliation functions are provided. In some examples, a system may receive data, from one or more computing systems. In some examples, one or more machine learning datasets may be used to identify datasets, data elements, or the like, for comparison. The identified datasets, data elements, and the like, may be compared to pre-stored patterns to determine whether the pattern matches a pre-stored pattern. If not, the pattern may be flagged as a new pattern and instructions for further processing may be requested. In some arrangements, the identified datasets, data elements, or the like, may be compared to determine whether a pattern and/or value of the datasets, data elements, or the like, matches. If not, one or more machine learning datasets may be used to generate a corrective action to align the data. In some examples, the generated corrective action may be automatically executed to align the data.

Abstract: Certain aspects of the present disclosure provide techniques for providing automated intelligence in a support session. In one example, a method includes generating a set of tokens based on a text-based query posted by a support agent to a live chat thread; generating a set of vectors based on the set of tokens; extracting a set of features based on the set of tokens; generating a query vector based on the set of vectors and the set of features; determining a predicted intent of the text-based query based on the query vector, wherein the predicted intent is one of a plurality of predefined intents; determining a predicted answer to the text-based query based on: the query vector; and the predicted intent; and providing the predicted answer to the text-based query in the live chat thread.

Abstract: Tuning a neural network may include selecting a portion of a first neural network for modification to increase computational efficiency and generating, using a processor, a second neural network based upon the first neural network by modifying the selected portion of the first neural network while offline.

Abstract: According to one embodiment, a computer-implemented method for active, imitation learning, includes: providing training data comprising an expert trajectory to a processor; querying the expert trajectory during an iterative, active learning process; generating a decision policy based at least in part on the expert trajectory and a result of querying the expert trajectory; attempting to distinguish the decision policy from the expert trajectory; in response to distinguishing the decision policy from the expert trajectory, outputting a policy update and generating a new decision policy based at least in part on the policy update; and in response to not distinguishing the decision policy from the expert trajectory, outputting the decision policy. Importantly, the expert trajectory is queried for only a subset of iterations of the iterative, active learning process, wherein the most uncertain state/action pair(s) from the expert trajectory are determined using one or more disagreement functions.

Abstract: A computer-implemented method includes employing a dynamic Boltzmann machine (DyBM) to predict a higher-order moment of time-series datasets. The method further includes acquiring the time-series datasets transmitted from a source node to a destination node of a neural network including a plurality of nodes, learning, by the processor, a time-series generative model based on the DyBM with eligibility traces, and obtaining, by the processor, parameters of a generalized auto-regressive heteroscedasticity (GARCH) model to predict a time-varying second-order moment of the times-series datasets.

Abstract: A neural network apparatus includes a plurality of node buffers connected to a node lane and configured to store input node data by a predetermined bit size; a plurality of weight buffers connected to a weight lane and configured to store weights; and one or more processors configured to: generate first and second split data by splitting the input node data by the predetermined bit size, store the first and second split data in the node buffers, output the first split data to an operation circuit for a neural network operation on an index-by-index basis, shift the second split data, and output the second split data to the operation circuit on the index-by-index basis.

Abstract: Embodiments of a system and method for identifying and prioritizing company prospects by training at least one classifier on client company win/loss metrics. One or more classifiers can be trained on a training database compiled from company win/loss database for a client and firmographic data from a robust business entity database. Once trained, the system can employ Artificial Intelligence powered by the trained classifiers to classify and output customized prospect lists of thousands of profiled and scored companies that the AI has determined are likely targets for specific marketing and sales. The AI can also ingest databases of client targets and classify and score them based on the custom-trained classifier.

Abstract: Disclosed are systems and methods for training and executing a neural network for collaborative monitoring of resource usage metrics. For example, a method may include receiving user data sets, grouping the user data sets into one or more clusters of user data sets, grouping each of the one or more clusters into a plurality of subclusters, for each of the plurality of subclusters, training the neural network to associate the subcluster with one or more sequential patterns found within the subcluster, grouping the plurality of user data sets into a plurality of teams, receiving a first series of transactions of a first user, inputting the first series of transactions into the trained neural network, classifying, using the trained neural network, the first user into a subcluster among the plurality of subclusters, generating a metric associated with the first series of transactions, generating a recommendation to the first user.

Abstract: There are provided system and method of classifying defects in a semiconductor specimen. The method comprises: upon obtaining by a computer a Deep Neural Network (DNN) trained to provide classification-related attributes enabling minimal defect classification error, processing a fabrication process (FP) sample using the obtained trained DNN; and, resulting from the processing, obtaining by the computer classification-related attributes characterizing the at least one defect to be classified, thereby enabling automated classification, in accordance with the obtained classification-related attributes, of the at least one defect presented in the FP image.

Abstract: A synapse system of a neuromorphic device may include a pre-synaptic neuron; a pre-synaptic line extending from the pre-synaptic neuron in a first direction; a post-synaptic neuron; a post-synaptic line extending from the post-synaptic line in a second direction; a selecting controller; a selecting line extending from the selecting controller in a third direction; and a synapse electrically connected with the pre-synaptic line, the post-synaptic line, and the selecting line.

Abstract: Techniques are disclosed for debugging a neural network execution on a target processor. A reference processor may generate a plurality of first reference tensors for the neural network. The neural network may be repeatedly reduced to produce a plurality of lengths. For each of the lengths, a compiler converts the neural network into first machine instructions, the target processor executes the first machine instructions to generate a first device tensor, and the debugger program determines whether the first device tensor matches a first reference tensor. A shortest length is identified for which the first device tensor does not match the first reference tensor. Tensor output is enabled for a lower-level intermediate representation of the shortest neural network, and the neural network is converted into second machine instructions, which are executed by the target processor to generate a second device tensor.

Abstract: Methods, systems, apparatuses, and computer program products are provided for a two-phase technique for generating content recommendations. In a first phase, a baseline recommender is configured to generate a baseline content recommendation using one or more content recommendation models, such as a Smart Adaptive Recommendations (SAR) model, Factorization Machine (FM) or Matrix Factorization (MF) models, collaborative filtering models, and/or any other machine-learning models or techniques. In a second phase, a personalized recommender implements a vector combiner configured to combine profile vectors, content vectors, and the baseline content recommendations to generate combined user vectors. A model generator may train a machine-learning model using the combined user vectors and training data comprising actual interaction behavior of the users, which may be then applied to identify a content recommendation for a particular user.

Abstract: A system predicts future positions or vergence depth of the user's eyes and generates gaze contingent content, such as for a head-mounted display (HMD), based on the predicted positions or vergence depth. The system includes an eye tracking controller that creates eye tracking information defining positions of a first eye and a second eye of a user over time. The eye tracking information is input to a neural network model that outputs the predicted positions or vergence depth. The predicted positions or vergence depth is then used to render the gaze contingent content, or to change other configurations of the HMD. Latency between the detection of user eye movement and the output of corresponding content is reduced to provide a more immersive real-time user experience.

Abstract: In one embodiment, a method includes accessing posts in a social-networking system. Each of the posts is unlabeled with respect to whether the post is known to be spam. The method also includes determining a posting user who submitted the post to the social-networking system and a recipient user to whom the post is addressed. The method further includes determining a first vector representation of the posting user and a second vector representation of the recipient user based on one or more features associated with the post, the posting user, and the recipient user. The method still further includes comparing the vector representations and building a machine learning model for automatically detecting spam posts in the social-networking system using a subset of the plurality of posts as non-spam training data.

Abstract: A device and a method for improving a processing speed of a neural network and applications thereof in the neural network where the device includes a processor configured to perform: determining, according to a predetermined processing speed improvement target, a dimension reduction amount of each of one or more parameter matrixes in the neural network obtained through training; preprocessing each parameter matrix based on the dimension reduction amount of the parameter matrix; and retraining the neural network based on a result of the preprocessing to obtain one or more dimension reduced parameter matrixes so as to ensure performance of the neural network meets a predetermined requirement. According to the embodiments of the present disclosure, it is possible to significantly improve the processing speed of the neural network while ensuring the performance of the neural network meets the predetermined requirement.