Abstract: A circuit for a multistage sequential data process includes a plurality of memory units. Each memory unit is associated with a stage of the sequential data process which, for each data set inputted to the stage, the stage provides an intermediate data set for storage in the associated memory unit for use in at least one subsequent stage of the sequential data process, where each of the plurality of memory units is sized based on relative locations of the stage providing the intermediate data set and the at least one subsequent stage in the sequential data process.

Abstract: Embodiments of a multimodal question answering (mQA) system are presented to answer a question about the content of an image. In embodiments, the model comprises four components: a Long Short-Term Memory (LSTM) component to extract the question representation; a Convolutional Neural Network (CNN) component to extract the visual representation; an LSTM component for storing the linguistic context in an answer, and a fusing component to combine the information from the first three components and generate the answer. A Freestyle Multilingual Image Question Answering (FM-IQA) dataset was constructed to train and evaluate embodiments of the mQA model. The quality of the generated answers of the mQA model on this dataset is evaluated by human judges through a Turing Test.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for receiving a request from a client to process a computational graph; obtaining data representing the computational graph, the computational graph comprising a plurality of nodes and directed edges, wherein each node represents a respective operation, wherein each directed edge connects a respective first node to a respective second node that represents an operation that receives, as input, an output of an operation represented by the respective first node; identifying a plurality of available devices for performing the requested operation; partitioning the computational graph into a plurality of subgraphs, each subgraph comprising one or more nodes in the computational graph; and assigning, for each subgraph, the operations represented by the one or more nodes in the subgraph to a respective available device in the plurality of available devices for operation.

Abstract: The disclosed embodiments illustrate method and system of processing data by a computing device for training a target domain classifier. The method includes extracting one or more first features and one or more second features from a first target instance, associated with a target domain. The method further includes predicting a first label for the received first target instance based on the one or more first features by utilizing a trained first classifier associated with a set of labeled source instances, wherein the predicted first label is assigned to the first target instance when a first score of the predicted first label exceeds a first pre-specified threshold. Further, the method includes updating a set of labeled target instances associated with the target domain based on the labeled first target instance, wherein the updated set of labeled target instances is utilized to train the target domain classifier.

Abstract: One or more processors receive hyperspectral band input, biometric input, and cognitive input as response input, from a user sampling a plurality of base foods, each base food prepared with a subset of ingredients and preparation techniques. The response input is transformed to a numeric representation of the respective input. Deep learning techniques are used to train an algorithm using the response data. A probabilistic ranking of base food is generated using unsupervised learning. Probability values of base food, ingredients, and preparation technique, associations preferred by the user, are generated, along with rules which define constraints associated with conditions for base food, ingredient, and preparation techniques, of user preferences. An objective function is generated that includes decision variables respectively aligned with constraints, and in response to optimizing the objective function, a preferred base food and ingredients, with preferred conditions of the user, is determined.

Abstract: A hybrid computer comprising a quantum processor can be operated to perform a scalable comparison of high-entropy samplers. Performing a scalable comparison of high-entropy samplers can include comparing entropy and KL divergence of post-processed samplers. A hybrid computer comprising a quantum processor generates samples for machine learning. The quantum processor is trained by matching data statistics to statistics of the quantum processor. The quantum processor is tuned to match moments of the data.

Abstract: A method and apparatus for processing data. The data is sent to a processor unit comprising a group of neural cores, a group of digital processing cores, and a routing network connecting the group of digital processing cores. The data is processed in the processor unit to generate a result.

Abstract: A method for training a question answering system includes providing training questions to a question answering system executing on a computer and to a plurality of subject matter experts. The question answering system generates first answers to each training question. Second answers to the training questions are received from the subject matter experts. Feature scores for each of the first answers and the second answers are generated and compared across the second answers and the first answers. Each of the feature scores is representative of a quality of an answer that is indicative of relevance to a corresponding training question. Based on the comparing, a measure of consistency of the feature scores of the second answers is determined, and a measure of consistency of the feature scores of the second answers to the first answers is determined. The measures of consistency are transmitted to the subject matter experts.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a policy neural network used to select actions to be performed by a reinforcement learning agent interacting with an environment. In one aspect, a method includes obtaining path data defining a path through the environment traversed by the agent. A consistency error is determined for the path from a combined reward, first and last soft-max state values, and a path likelihood. A value update for the current values of the policy neural network parameters is determined from at least the consistency error. The value update is used to adjust the current values of the policy neural network parameters.

Abstract: In one embodiment, a server in a network reports one or more symptoms of a monitored device that is malfunctioning to a user interface via a particular chatbot session. The server receives, via the particular chatbot session, a triage request to enter a triage mode regarding the one or more reported symptoms. The server predicts a corrective action using the one or more reported symptoms as input to a machine learning model. The machine learning model is trained using a history of observed symptoms in the network, a history of corrective actions initiated via chatbot sessions and associated with the observed symptoms, and a history of feedback regarding the corrective actions received via the chatbot sessions. The server provides the predicted corrective action to the user interface via the particular chatbot session as a suggested corrective action, in response to the received triage request.

Abstract: There are disclosed devices, system and methods for a machine learning binary classifier automatically tolerating training data that is incorrect by determining a correct and an incorrect likelihood ratio that each training data entry has a correctly and an incorrectly labeled output. The correct and an incorrect likelihood ratio are combined with a correct and an incorrect priori odds ratio that the set of training data entries have correctly and incorrect labeled output labels. These two combinations are a correct probability and an incorrect probability that each entry of the set of entries has a correctly and an incorrect labeled output. A logistic regression model if fit to a combination of the correct probability and the incorrect probability for each training data entry to complete the training.

Abstract: The disclosed systems and methods include pre-calculation, per object, of object feature bin values, for identifying close matches between objects, such as text documents, that have numerous weighted features, such as specific-length word sequences. Predetermined feature weights get scaled with two or more selected adjacent scaling factors, and randomly rounded. The expanded set of weighted features of an object gets min-hashed into a predetermined number of feature bins. For each feature that qualifies to be inserted by min-hashing into a particular feature bin, and across successive feature bins, the expanded set of weighted features get min-hashed and circularly smeared into the predetermined number of feature bins. Completed pre-calculated sets of feature bin values for each scaling of the object, together with the scaling factor, are stored for use in comparing sampled features of the object with sampled features of other objects by calculating an estimated Jaccard similarity index.

Abstract: The present disclosure generally includes a projective modeling and simulation system which produces an Outcome Model which reflects the projection of a structured assertion across the elements composing a Reference Data Model, where the projected contention is embodied within a structured Assertion Model and an optionally associated but similarly structured Apportionment sub-Model where the Outcome Model unifies the subject matter of the Reference Data Model with the Assertion-Apportionment Model pair.

Abstract: Generating a solution keyword tag cloud is provided. Keywords are identified in a question asking how to resolve an issue experienced by a user with a product. The keywords identified in the question are matched with keyword tags included in a set of condition-solution trees corresponding to the product. The solution keyword tag cloud for the product is generated based on the matching of the keywords identified in the question with the keyword tags included in the set of condition-solution trees corresponding to the product. The solution keyword tag cloud is presented in a graphical user interface display on a client device corresponding to the user.

Abstract: Provided are an apparatus and method for classifying home appliances based on power consumption using deep learning, which can efficiently classify home appliances in use by applying deep learning and analyzing power data collected from a house. The apparatus includes a home appliance classification model creation module configured to encode power consumption data collected from a house to learn a home appliance classification model and create an RNN-based home appliance classification model and a home appliance classification module configured to collect and encode data on power consumption currently in use and classify home appliances using the home appliance classification model created by the home appliance classification model creation module.