Abstract: In one embodiment, an apparatus comprises a storage device and a processor. The storage device stores a feature vector index, wherein the feature vector index comprises a sparse-array data structure representing a feature space for a set of labeled feature vectors, wherein the set of labeled feature vectors are assigned to a plurality of classes. The processor is to: receive a query corresponding to a target feature vector; access, via the storage device, a first portion of the feature vector index, wherein the first portion of the feature vector index comprises a subset of labeled feature vectors that correspond to a same portion of the feature space as the target feature vector; determine the corresponding class of the target feature vector based on the subset of labeled feature vectors; and provide a response to the query based on the corresponding class.

Abstract: A device includes an image capture device configured to capture a first video. The device includes a memory configured to store one or more videos. The device further includes a processor coupled to the memory. The processor is configured to concatenate the first video and a second video to generate a combined video. The second video is included in the one or more videos or is accessible via a network. The second video is selected by the processor based on a similarity of a first set of characteristics with a second set of characteristics. The first set of characteristics corresponds to the first video. The second set of characteristics corresponds to the second video.

Abstract: A data hub for servicing data hub dependent data consumers includes a persistent storage and a data validator. The persistent storage stores validated data. The data validator obtains a data validation request; in response to obtaining the data validation request: imports data from a data aggregator to obtain the validated data; performs a continuity analysis of the validated data to generate a data deviation report; and provides a portion of the validated data to one of the data hub dependent data consumers.

Abstract: A method and a system for training a machine learning algorithm (MLA) for object classification. The machine learning algorithm includes an embedding layer and a classification layer. A set of embedding indices representing a reference object is received. The set of embedding indices has been generated based on a byte representation of the reference object. A label associated with the reference object indicative of a reference class the objects belongs to is received. The MLA is iteratively trained to classify objects by embedding the set of embedding indices to obtain an input vector and by predicting an estimated class based on the input vector, and updating a parameter of at least one of the embedding layer and the updated embedding layer. The set of embedding indices is generated by parsing the byte representation to obtain byte n-grams and by applying a hash function on the byte n-grams.

Abstract: A computer implemented method and system for a selectively replicated trustless persistent store is provided using a bilateral distributed ledger. The selectively replicated trustless persistent store synchronizes current state data stores shared among multiple parties. Data modifications may be made in any shared store locally and then are automatically replicated across other permissioned stores. The selectively replicated trustless persistent store is responsible for getting the data validated and agreed upon before committing locally.

Abstract: Methods and systems for optimizing performance of a cyber-physical system include training a machine learning model, according to sensor data from the cyber-physical system, to generate one or more parameters for controllable sensors in the cyber-physical system that optimize a performance indicator. New sensor data is collected from the cyber-physical system. One or more parameters for the controllable sensors are generated using the trained machine learning module and the new sensor data. The one or more parameters are applied to the controllable sensors to optimize the performance of the cyber-physical system.

Abstract: Embodiments described herein provide techniques for maintaining consistency in a distributed system (e.g., a distributed secondary storage system). According to one embodiment of the present disclosure, a first set of file system objects included in performing the requested file system operation is identified in response to a request to perform a file system operation. An update intent corresponding to the requested file system operation is inserted into an inode associated with each identified file system object. Each file system object corresponding to the inode is modified as specified by the update intent in that inode. After modifying the file system object corresponding to the inode, the update intent is removed from that inode.

Abstract: An approach is provided for completing a pipeline graph. Using a deep learning based sequence model, an initial data pipeline having a sequence of nodes is generated. Mismatch(es) between data formats required by input and output in the sequence of nodes is identified. Virtual gap node(s) that correct the mismatch(es) are added to the initial data pipeline. For a given virtual gap node, tentative graph structures are determined using knowledge graphs and a crowd sourced validation system. Reuse forecast scores and performance scores for the tentative graph structures are calculated. Based on the reuse forecast scores and the performance scores, a final graph structure for implementing the given virtual gap node is determined.

Abstract: The invention provides a neural network quantization method and device and a related product. The neural network quantization method is used for quantizing data of a computation layer of a neural network. The technical scheme provided by the invention has the advantage of low cost.

Abstract: The subject technology executes a change on an existing micro-partition of a table of a database, the executing of the change comprising generating a new micro-partition that embodies the change. The subject technology receives a request for a delta for the table between a first timestamp and a second timestamp. The subject technology queries at least one change tracking column to determine the delta between the first timestamp and the second timestamp, the delta including information indicating at least one database operation that was performed to at least one row of a set of rows of the table, without including information as to intermediate changes made to at least one row of the set of rows of the table between the first timestamp and the second timestamp, that facilitates a reduction in storage of historical versions of the set of rows of the table.

Abstract: An embodiment includes identifying metadata attached to a first node of a graph, where the metadata satisfies an ontological condition. The embodiment also includes transforming the graph such that the transforming results in the graph having a new graph structure, where the transforming of the graph comprises removing the metadata from the first node and adding a second node representative of the metadata removed from the first node to the graph such that the second node is connected to the first node by a first edge.

Abstract: A method creates an embedding for an unlabeled vertex in a hypergraph. The method includes receiving a hypergraph of hyperedges, where each of the hyperedges includes one or more vertices, and at least one of the hyperedges includes an unlabeled vertex; generating a hypergraph of vertices from the hypergraph of hyperedges, where each of the vertices in the hypergraph of vertices includes one or more of the one or more hyperedges from the hypergraph of hyperedges; performing a first type of random walk through the hypergraph of hyperedges; performing a second type of random walk through the hypergraph of vertices; generating a set of vertex embeddings from the first type of random walk and a set of hyperedge embeddings from the second type of random walk; and using results of the first and second random walks to train a neural network to create an embedding for the unlabeled vertex.

Abstract: A method for assigning weights to a knowledge graph includes extracting information from a knowledge graph. The information including entities extracted from nodes of the knowledge graph and relations extracted from edges of the knowledge graph. A shortest path generator receives the extracted entities and relations, and potential assigned weights from a heuristic data repository. Weights for the edges of the knowledge graph are determined. The weights are assigned to the edges of the knowledge graph.

Abstract: A method, a system, and a computer program product for fast training and/or execution of neural networks. A description of a neural network architecture is received. Based on the received description, a graph representation of the neural network architecture is generated. The graph representation includes one or more nodes connected by one or more connections. At least one connection is modified. Based on the generated graph representation, a new graph representation is generated using the modified at least one connection. The new graph representation has a small-world property. The new graph representation is transformed into a new neural network architecture.

Abstract: A method for graph data processing comprises obtaining graph data which includes a plurality of nodes and data corresponding to the plurality of nodes respectively; classifying the plurality of nodes into at least one category of a plurality of categories, wherein the plurality of categories are associated with a plurality of node relationship patterns; determining, from a plurality of candidate parameter value sets of a graph convolutional network (GCN) model, parameter value subsets respectively matching at least one category, wherein the plurality of candidate parameter value sets are determined by training the GCN model respectively for the plurality of node relationship patterns; and using the parameter value subsets respectively matching the at least one category to respectively perform a graph convolution operation in the GCN model on data corresponding to the nodes classified into the at least one category to obtain a processing result for the graph data.

Abstract: A data management system trains an analysis model with a machine learning process to understand the semantic meaning of queries received from users of the data management system. The machine learning process includes retrieving assistance documents that each include a query and an answer to the query. A training model analyzes each answer and generates first topic distribution data indicating, for each answer, how relevant each of a plurality of topics is to the answer. The queries are passed to the analysis model and the analysis model is trained to generate second topic distribution data that converges with the first topic distribution data based on analysis of the queries.

Abstract: An apparatus for goal generation is disclosed. The apparatus includes at least a processor and a memory communicatively connected to the at least a processor. The memory instructs the processor to receive a goal datum related to a user, wherein the goal datum comprises behavioral parameters. The memory additionally instructs the processor to classify the goal datum to a user goal. The classification comprises training a goal classifier using a goal training data. Goal training data contains a plurality of data entries containing a plurality of goal datum inputs correlated to a plurality of goal outputs. The classification also comprises classifying the goal datum to the goal using the goal classifier. The classifier assigns the goal as a function of the classification. A goal path is generated as a function of the classification of the goal datum to a goal, wherein the goal path is divided into waypoints.

Abstract: Methods, systems, and non-transitory computer readable storage media are disclosed for determining clusters of similar digital documents using unique document signatures. Specifically, the disclosed system processes digital text in a digital document to tokenize character strings (e.g., words) in the digital document by combining a subset of character values and string lengths in the character strings. Additionally, the disclosed system generates a document signature for the digital document by combining subsets of tokens generated for the digital document into a token sequence indicative of the digital text in the digital document. The disclosed system determines a cluster of similar digital documents including the digital document by comparing the document signature of the digital document to document signatures corresponding to a plurality of digital documents.

Abstract: One example method includes receiving, as an input, an ML pipeline definition, identifying a group of layers required to be created for the ML pipeline definition, for one of more of the layers, receiving input concerning one or more characteristics of the layer, creating the layers for which input has been received, and packaging the created layers with the ML pipeline definition to create a production-ready ML model.

Abstract: Systems and methods are disclosed for automatically detecting a design element in a design document. One method comprises receiving a design document and generating an enhanced design document based on the received design document. The enhanced design document may be generated by augmenting additional information to the design document using machine learning techniques. In response to receiving a user input, one or more design elements in the enhanced design document may be determined, and additional information associated with the determined one or more design elements may be displayed to the user.