Abstract: A computation unit for performing a computation of a neural network layer is disclosed. A number of processing element (PE) units are arranged in an array. First input values are provided in parallel in an input dimension of the array during a first processing period, and a second input values are provided in parallel in the input dimension during a second processing period. Computations are performed by the PE units based on stored weight values. An adder coupled to the first set of PE units generates a first sum of results of the computations by the first set of PE units during the first processing cycle, and generates a second sum of results of the computations during the second processing cycle. A first accumulator coupled to the first adder stores the first sum, and further shifts the first sum to a second accumulator prior to storing the second sum.

Abstract: A quantum state classifier includes a reservoir computing circuit for post-processing a quantum bit to obtain a readout signal, and a readout circuit, coupled to the reservoir computing circuit, for discriminating a quantum state of the quantum bit from the readout signal from among multiple possible quantum states. The readout circuit is trained in a calibration process respectively activated by a specific one of each of the multiple quantum states such that weights within the linear readout circuit are updated by minibatch learning for each of multiple measurement sequences of the calibration process. The readout circuit generates a binary output after the multiple measurement sequences in a post-calibration classification process for a test quantum bit. The quantum state classifier further includes a controller, coupled to the readout circuit, selectively triggerable to output a control pulse responsive to the quantum state of the test quantum bit indicated by the binary output.

Abstract: Memory cells can include a memory region to store a machine learning model and input data and another memory region to store host data from a host system. An in-memory logic can be coupled to the plurality of memory cells and can perform a machine learning operation by applying the machine learning model to the input data to generate an output data. A bus can receive additional host data from the host system and can provide the additional host data to the memory component for the other memory region of the plurality of memory cells. An additional bus can receive machine learning data from the host system and can provide the machine learning data to the memory component for the in-memory logic that is to perform the machine learning operation.

Abstract: Provided are a training apparatus, a recognition apparatus, a training method, a recognition method, and a program that can accurately recognize what an object represented in an image associated with depth information is. An object data acquiring section acquires three-dimensional data representing an object. A training data generating section generates a plurality of training data each representing a mutually different part of the object on the basis of the three-dimensional data. A training section trains a machine learning model using the generated training data as the training data for the object.

Abstract: Methods for training a neural sequence-to-sequence (seq2seq) model. A processor receives the model and training data comprising a plurality of training source sequences and corresponding training target sequences, and generates corresponding predicted target sequences. Model parameters are updated based on a comparison of predicted target sequences to training target sequences to reduce or minimize both a local loss in the predicted target sequences and an expected loss of one or more global or semantic features or constraints between the predicted target sequences and the training target sequences given the training source sequences. Expected loss is based on global or semantic features or constraints of general target sequences given general source sequences.

Abstract: Provided are an AI system for simulating functions such as recognition, determination, and so forth of human brains by using a mechanical learning algorithm like deep learning, or the like, and an application thereof. In particular, according to the AI system and the application thereof, a neural network training method includes obtaining a plurality of first images belonging to a particular category and a plurality of second images for which a category is not specified, training a neural network model for category recognition, based on the plurality of first images belonging to the particular category, recognizing at least one second image corresponding to the particular category among the plurality of second images, by using the trained neural network model, and modifying and refining the trained neural network model based on the recognized at least one second image.

Abstract: A method of updating a neural network model by a terminal device, includes training a local model using a local data set collected by a terminal device to generate a trained local model; receiving, from a server, an independent identically distributed (i.i.d.) global data set, the i.i.d. global data set being a data set sampled for each class in a plurality of predefined classes; implementing the trained local model by inputting the i.i.d. global data set and transmitting final inference results of the implemented trained local model to the server; and receiving, from the server, a global model updated based on the final inference results of the inference.

Abstract: Techniques regarding multivariate time series data analysis are provided. For example, one or more embodiments described herein can comprise a system, which can comprise a memory that can store computer executable components. The system can also comprise a processor, operably coupled to the memory, and that can execute the computer executable components stored in the memory. The computer executable components can comprise a time series analysis component that generates a machine learning model that discovers a dependency between multivariate time series data using an attention mechanism controlled by an uncertainty measure.

Abstract: A processor-implemented method of performing a convolution operation is provided. The method includes obtaining input feature map data and kernel data, determine the kernel data based on a number of input channels of the input feature map, a number of output channels of an output feature map, and a number of groups of the input feature map data and a number of groups of the kernel data related to the convolution operation, and performing the convolution operation based on the input feature map data and the determined kernel data.

Abstract: A system maintains a knowledge layout to support the building of event and analytics models in parity. The system uses the event models to provide a snapshot of the relevant conditions present when a challenge event occurs. The system uses the analytics models to select one or more actions (which may include robotic tasks) to respond to the challenge condition. In some cases, the system may render continued response compulsory until a successful response to the challenge event is achieved.

Abstract: Systems and methods for training a neural network or an ensemble of neural networks are described. A hyper-parameter that controls the variance of the ensemble predictors is used to address overfitting. For larger values of the hyper-parameter, the predictions from the ensemble have more variance, so there is less overfitting. This technique can be applied to ensemble learning with various cost functions, structures and parameter sharing. A cost function is provided and a set of techniques for learning are described.

Abstract: Mechanisms are provided to provide an improved computer tool for determining and mitigating the presence of adversarial inputs to an image classification computing model. A machine learning computer model processes input data representing a first image to generate a first classification output. A cohort of second image(s), that are visually similar to the first image, is generated based on a comparison of visual characteristics of the first image to visual characteristics of images in an image repository. A cohort-based machine learning computer model processes the cohort of second image(s) to generate a second classification output and the first classification output is compared to the second classification output to determine if the first image is an adversarial image. In response to the first image being determined to be an adversarial image, a mitigation operation by a mitigation system is initiated.

Abstract: The present disclosure relates to an object selection system that automatically detects and selects objects in a digital image utilizing a large-scale object detector. For instance, in response to receiving a request to automatically select a query object with an unknown object class in a digital image, the object selection system can utilize a large-scale object detector to detect potential objects in the image, filter out one or more potential objects, and label the remaining potential objects in the image to detect the query object. In some implementations, the large-scale object detector utilizes a region proposal model, a concept mask model, and an auto tagging model to automatically detect objects in the digital image.

Abstract: Embodiments of the present disclosure disclose a method and apparatus for compressing a deep learning model. An embodiment of the method includes: acquiring a to-be-compressed deep learning model; pruning each layer of weights of the to-be-compressed deep learning model in units of channels to obtain a compressed deep learning model; and sending the compressed deep learning model to a terminal device, so that the terminal device stores the compressed deep learning model. By pruning each layer of weights of the deep learning model in units of channels, the parameter redundancy of the deep learning model is effectively reduced, thereby improving the computational speed of the deep learning model and maintaining the model accuracy.

Abstract: A method of generating a second neural network model according to an example embodiment includes: inputting unlabeled input data to a first neural network model; obtaining prediction results corresponding to the unlabeled input data based on the first neural network model; and generating a second neural network model based on the prediction results of the first neural network model and a degree of distribution of the prediction results.

Abstract: An inference system trains and performs inference using a sparse neural network. The sparse neural network may include one or more layers, and each layer may be associated with a set of sparse weights that represent sparse connections between nodes of a layer and nodes of a previous layer. A layer output may be generated by applying the set of sparse weights associated with the layer to the layer output of a previous layer. Moreover, the one or more layers of the sparse neural network may generate sparse layer outputs. By using sparse representations of weights and layer outputs, robustness and stability of the neural network can be significantly improved, while maintaining competitive accuracy.

Abstract: Intent determination based on one or more multi-model structures can include generating an output from each of a plurality of domain-specific models in response to a received input. The domain-specific models can comprise simultaneously trained machine learning models that are trained using a corresponding local loss metric for each domain-specific model and a global loss metric for the plurality of domain-specific models. The presence or absence of an intent corresponding to one or more domain-specific models can be determined by classifying the output of each domain-specific model.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network. One of the methods includes receiving training data; training a neural network on the training data, wherein the neural network is configured to: receive a network input, convert the network input into a latent representation of the network input, and process the latent representation to generate a network output from the network input, and wherein training the neural network on the training data comprises training the neural network on a variational information bottleneck objective that encourages, for each training input, the latent representation generated for the training input to have low mutual information with the training input while the network output generated for the training input has high mutual information with the target output for the training input.

Abstract: As described, an artificial intelligence (AI) design application exposes various tools to a user for generating, analyzing, evaluating, and describing neural networks. The AI design application includes a network generator that generates and/or updates program code that defines a neural network based on user interactions with a graphical depiction of the network architecture. The AI design application also includes a network analyzer that analyzes the behavior of the neural network at the layer level, neuron level, and weight level in response to test inputs. The AI design application further includes a network evaluator that performs a comprehensive evaluation of the neural network across a range of sample of training data. Finally, the AI design application includes a network descriptor that articulates the behavior of the neural network in natural language and constrains that behavior according to a set of rules.

Abstract: According to one aspect of the present invention, a learning device is configured to perform learning of decision trees by gradient boosting. The learning device includes an initializer and a learning unit. The initializer is configured to perform initialization processing on an address memory that stores an address in a data memory of learning data used for learning of a node in a decision tree. The learning unit is configured to perform learning of the decision tree by using the learning data stored in the data memory. The initializer is configured to perform, after a certain decision tree is learned by the learning unit, the initialization processing for a next decision tree of the certain decision tree in parallel with processing for the certain decision tree.

Abstract: A controller generating a knowledge graph of entries, each entry comprising a separate entity identifier and a separate entity mention identifier within a separate document of a corpus of documents with a located relationship and one or more computed prefix-based geotemporal values determined from geotemporal information associated with the separate entity mention identifier within the separate document. The controller, in response to receiving an input comprising a particular entity and a threshold value, mapping the threshold value to a geospatial hash prefix type and a temporal hash prefix type. The controller applying geospatial hash prefix type and the temporal hash prefix type to the entries in the knowledge graph to determine a response to the input indicating one or more geotemporal proximate entities identified within a degree of geotemporal proximity to the particular entity set by the threshold value.

Abstract: Systems, computer-implemented methods, and computer program products that can facilitate refinement of a predicted event based on explainability data are provided. According to an embodiment, a system can comprise a memory that stores computer executable components and a processor that executes the computer executable components stored in the memory. The computer executable components can comprise an interpreter component that identifies a probable cause of a predicted event based on explainability data. The computer executable components can further comprise an enrichment component that executes a diagnostic analysis based on the probable cause.

Abstract: Systems and methods are disclosed for predicting a remaining useful life of a component. One method comprises receiving, by a component prediction system, one or more component data sets associated with one or more components of a moving object. Based on the received one or more component data sets, environmental and operational conditions experienced over a lifetime of each failed component may be identified and summarized. Then, the effects of the environmental and operational conditions on a lifetime of a component of interest may be determined by training an accelerated failure time model using the summarized environmental and operational conditions. Using the trained accelerated failure time model, a remaining useful life of the component of interest may be determined.

Abstract: The present disclosure relates to a method for enabling data integration. The method comprises collecting matching results of matching of records by a matching component over a time window. The number of false tasks of user defined tasks and system defined tasks in the collected matching results may be determined. The matching criterion used by the matching component may be adjusted to minimize the number of user defined tasks while the fraction of false tasks stays within a certain limit. The matching criterion may be replaced by the adjusted matching criterion for further usage of the matching component.

Abstract: A system that provides a mathematical formulation for new problem of model validation and model selection in presence of test data feedback. The system comprises a memory that stores computer-executable components. A processor, operably coupled to the memory, executes the computer-executable components stored in the memory. A selection component selects a metric of performance evaluation accuracy; and a configuration component configures performance evaluation schemes for machine learning algorithms. A characterization component employs a supervised learning-based approach to characterize relationship between the configuration of the performance evaluation scheme and fidelity of performance estimates; and an optimization component that optimizes accuracy of the machine learning algorithms as a function of size of training data set relative to size of validation data set through selection of values associated with the configuration parameters.

Abstract: A first and second blending profile may be created for a set of question answering pipelines. A set of test answer data may be generated for a first answering pipeline. The test answer data may be generated based on a set of test question and using an answer key associated with the test questions. Based on the test answer data, a first blending profile can be created for the first answering pipeline. Using the set of test questions and a second answer key, another set of test answer data may be generated. This set may be generated for the second answering pipeline. Using this second answering pipeline test answer data, a second blending profile can be generated for the second answering pipeline. Each blending profile may have metadata about a confidence of each pipeline.

Abstract: Methods and apparatus for improving automatic selection and timing of messages by a machine or system of machines include an inductive computational process driven by log-level network data from mobile devices and other network-connected devices, optionally in addition to traditional application-level data from cookies or the like. The methods and apparatus may be used, for example, to improve or optimize effectiveness of automatically-generated electronic communications with consumers and potential consumers for achieving a specified target.

Abstract: A computer implemented method for testing rules by a computing device including selecting a current version of a rule and prior version of the rule, comparing the prior version of the rule and the current version of the rule to each other to identify a type of change made in the current version of the rule with respect to the prior version of the rule, and testing the prior version of a rule and the current version of the rule using a common data set, the testing being based on the identified type of change. The test result is provided to a user.

Abstract: A method performed by a data analysis apparatus according to an embodiment of the present disclosure includes generating a plurality of module combination processes using a plurality of data analysis modules defined by a user, calculating a score for each of the data analysis modules based on an execution result of the plurality of module combination processes and generating a recommendation module candidate group including a combination of data analysis modules selected based on the score.

Abstract: Systems and methods are disclosed to infer, using a machine learned model, a service protocol of a server based on the banner data produced by the server. In embodiments, the machine learned model is implemented by a network scanner configured to receive banner data from open ports on servers. A received banner is parsed into a set of features, such as the counts or presence of particular characters or strings in the banner. In embodiments, certain types of banner content such as network addresses, hostnames, dates, and times, are replaced with special characters. The machine learned model is applied to the features to infer a most likely protocol of the server port that produced the banner. Advantageously, the model can be trained to perform the inference task with high accuracy and without using human-specified rules, which can be brittle for unconventional banner data and carry undesired biases.

Abstract: A system, method, and electronic online platform provide a probability of a wager winning in real-time updates as live in-game data is provided to a model calculating a current probability of winning based on historical betting data.

Abstract: This disclosure is directed to an apparatus for intelligent matching of disparate input data received from disparate input data systems in a complex computing network for establishing targeted communication to a computing device associated with the intelligently matched disparate input data.

Abstract: This disclosure relates generally to the field of quantum algorithms and quantum data loading, and more particularly to constructing quantum circuits for loading classical data into quantum states which reduces the computational resources of the circuit, e.g., number of qubits, depth of quantum circuit, and type of gates in the circuit.

Abstract: Degeneracy in analog processor (e.g., quantum processor) operation is mitigated via use of floppy qubits or domains of floppy qubits (i.e., qubit(s) for which the state can be flipped with no change in energy), which can significantly boost hardware performance on certain problems, as well as improve hardware performance for more general problem sets. Samples are drawn from an analog processor, and devices comprising the analog processor evaluated for floppiness. A normalized floppiness metric is calculated, and an offset added to advance the device in annealing. Degeneracy in a hybrid computing system that comprises a quantum processor is mitigated by determining a magnetic susceptibility of a qubit, and tuning a tunneling rate for the qubit based on a tunneling rate offset determined based on the magnetic susceptibility. Quantum annealing evolution is controlled by causing the evolution to pause for a determined pause duration.

Abstract: The present disclosure describes non-classical (e.g., quantum) computing systems and methods that utilize dopant molecules contained in host materials as qubits.

Abstract: One or more embodiments of a content naming system provide machine-learned name suggestions to a user for naming content items. Specifically, an online content management system can train a machine-learning model to identify a naming pattern from previously stored content items corresponding to a user account of the user. The online content management system uses the machine-learning model to determine a plurality of name suggestions for naming a content item associated with the user account. One or more embodiments provide graphical elements corresponding to the name suggestions within a graphical user interface. The user can select one or more graphical elements to add the corresponding name suggestion(s) to the name of the content item.

Abstract: In some embodiments, user-selectable/connectable model representations may be provided via a user interface to facilitate artificial intelligence development. The model representations may comprises first and second machine learning model (ML) representations corresponding to first and second ML models, and non-ML model representations corresponding to non-ML models. Based on user input indicating selection of the first and second ML model representations and a non-ML model representation corresponding to a non-ML model, at least a portion of a software application may be generated such that the software application comprises (i) an instance of the first ML model, an instance of the second ML model, and an instance of the non-ML model and (ii) an input/output data path between the instance of the first ML model and at least one other instance, the at least one other instance comprising the instance of the second ML model or the instance of the non-ML model.

Abstract: “Semi-supervised” machine learning relies on less human input than a supervised algorithm to train a machine learning algorithm to perform entity recognition (NER). Starting with a known entity value or known pattern value for a specific entity type, phrases in a training data corpus are identified that include the known entity value. Context-value patterns are generated to match selected phrases that include the known entity value. One or more context-value patterns may be validated based on human input. The validated patterns identify additional entity values. A subset of the additional entity values may also be validated based on human input. Occurrences of validated entity values may be labeled in the training corpus. Sample phrases from the labeled training dataset may be extracted to form a reduced-size training set for a supervised machine learning model which may be further used in production to label data for any named entity recognition application.

Abstract: The disclosed technology relates to a process for metered training of fog nodes within the fog layer. The metered training allows the fog nodes to be continually trained within the fog layer without the need for the cloud. Furthermore, the metered training allows the fog node to operate normally as the training is performed only when spare resources are available at the fog node. The disclosed technology also relates to a process of sharing better trained machine learning models of a fog node with other similar fog nodes thereby speeding up the training process for other fog nodes within the fog layer.

Abstract: Methods, systems, and computer-readable storage media for determining, by an automated regression detection system (ARDS), that training of a ML model is complete, the ML model being a version of a previously trained ML model, and in response, automatically, by the ARDS: retrieving the ML model, executing regression testing and detection using the ML model, generating regression results relative to the previously trained ML model, and publishing the regression results.

Abstract: A method of selecting a model of machine learning executed by a processor is provided. The method includes: receiving at least one data-set; configuring a configuration space for machine learning of the at least one data-set; extracting, from the at least one data-set, a meta-feature including quantitative information about the data-set; calculating performance of the machine learning for the at least one data-set based on a plurality of configurations included in the configuration space; executing meta-learning based on the meta-feature, the plurality of configurations, and the calculated performance; and optimizing the configuration space based on a result of executing the meta-learning.

Abstract: The system described herein may utilize a data stream connection to detect that a new message is transmitted between users. The system may perform various pre-processing techniques on the new message to identify that the new message is an objection message candidate. The system may retrieve one or parent messages of the new message. The new message and the parent messages may be input into a model trained to classify objection messages. The model may identify that the new message is classified as an objection message based at least in part on processing the new message and the parent messages. An objection classification identifier may be stored in association with the new message based at least in part on the result of the processing by the model.

Abstract: A computer-implemented method includes: learning, by a computer device, a delivery for response content in view of types of queries; awakening, by the computer device, in response to receiving an activation command; receiving, by the computer device, a query; determining, by the computer device, a context of the query; determining, by the computer device, a digestibility of a response to the query; and determining, by the computing device, to output a response to the query as one of an audio response, a displayed response, and an audio response and a displayed response to the user, wherein the determining is based on the learning, the determined context of the query, the determined digestibility of the response, and the preferences of the user for receiving the response.

Abstract: A method for identifying a scene, comprising a computing device receiving a plurality of data points corresponding to a scene; the computing device determining one or more subsets of data points from the plurality of data points that are indicative of at least one sub-scene in said scene, said at least one sub-scene displayed on a display device that is part of said scene, wherein said at least one sub-scene does not represent said scene; the computing device categorizing said scene, disregarding said at least one sub-scene, wherein the categorizing includes interpreting said scene by a computer vision system such that said at least one sub-scene is not taken into account in the categorizing of said scene.

Abstract: A method, a computer system, and a computer program product are provided for federated learning. An aggregator may receive cluster information from distributed computing devices. The cluster information may relate to identified clusters in sample data of the distributed computing devices. The cluster information may include centroid information per cluster. The aggregator may include a processor. The aggregator may integrate the cluster information to define data classes for machine learning classification. The integrating may include computing a respective distance between centroids of the clusters in order to determine a total number of the data classes. The aggregator may send a deep learning model that includes an output layer that has a total number of nodes equal to the total number of the data classes. The deep learning model may be for the distributed computing devices to perform machine learning classification in federated learning.

Abstract: A system and method is disclosed for augmenting image data of an invasive medical device using optical imaging. An optical imaging sensor, separate from the invasive medical device, can generate images of the medical device within a patient. A trained model for the invasive medical device can be trained on annotated images of the invasive medical device with orientation and distance information of the invasive medical device. An imaging computer system can apply the trained model to images of the invasive medical device within the patient to determine a current orientation and a current distance of the invasive medical device. The images of the invasive medical device as captured by the optical imaging sensor, visual orientation information representing the current orientation of the invasive medical device, and visual distance information representing the current distance of the invasive medical device within the patient can be displayed.

Abstract: Systems and methods that analyze blood-based cancer diagnostic tests using multiple classes of molecules are described. The system uses machine learning (ML) to analyze multiple analytes, for example cell-free DNA, cell-free microRNA, and circulating proteins, from a biological sample. The system can use multiple assays, e.g., whole-genome sequencing, whole-genome bisulfite sequencing or EM-seq, small-RNA sequencing, and quantitative immunoassay. This can increase the sensitivity and specificity of diagnostics by exploiting independent information between signals. During operation, the system receives a biological sample, and separates a plurality of molecule classes from the sample. For a plurality of assays, the system identifies feature sets to input to a machine learning model. The system performs an assay on each molecule class and forms a feature vector from the measured values.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing parallel generation of output from an autoregressive sequence to sequence model. In one aspect, a blockwise parallel decoding method takes advantage of the fact that some architectures can score sequences in sublinear time. By generating predictions for multiple time steps at once then backing off to a longest prefix validated by the scoring model, the methods can substantially improve the speed of greedy decoding without compromising performance.

Abstract: The disclosed computer-implemented method may include instant and optimized matching of transportation requesters with transportation providers by precomputing and caching evaluations of matching schemes between existing requests and available providers, where each of the matching schemes excludes one of the available providers. The possibility of matching any new request to a given provider may then be evaluated according to the cached results of the matching scheme that excluded that provider along with an evaluation of matching the new request to the provider, which may be a computationally easy problem. Thus, the new request may be matched to a provider instantly without waiting for the next iteration of solving the global matching problem. Matching requestors and providers in this way may also improve the accuracy of estimated time of arrival information provided to requestors. Various other methods, systems, and computer-readable media are also disclosed.

Abstract: A system and method for selecting inventory pricing for an event at a venue is disclosed. The method comprising determining a rate at which a first inventory of seats have sold for an event at a venue. The method further comprising calculating a demand for a secondary inventory of seats as a function of the rate at which the first inventory of seats sold, the seats of the first and second inventories being comparable in quality, provides a user interface to one or more client devices that displays the data. The method further comprising calculating a demand for a second inventory of seats as a function of the rate at which the first inventory of seats sold, the seats of the first and second inventories being comparable in quality.