Abstract: A client device determines a local user gradient value based on a current user preference vector and a local item gradient value based on a current item feature vector. The client device updates a user preference vector by using the local user gradient value and updates an item feature vector by using the local item gradient value. The client device determines a neighboring client device based on a predetermined adjacency relationship. The local item gradient value is sent by the client device to the neighboring client device. The client device receives a neighboring item gradient value sent by the neighboring client device. The client device updates the item feature vector by using the neighboring item gradient value. In response to the client device determining that a predetermined iteration stop condition is satisfied, the client device outputs the user preference vector and the item feature vector.

Abstract: A method for analysis of problems is described, comprising receiving values for a plurality of input parameters specifying a problem, analyzing the values of the plurality of input parameters with a fuzzy expert system thereby calculating a fuzzy result, including a value of a linguistic variable and a crisp value, and determining a priority of the problem, wherein the priority is determined based on the value of the linguistic variable and the crisp value of the fuzzy result. Furthermore, a corresponding problem analysis system is provided.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for executing a signal generation neural network on parallel processing hardware. One of the methods includes receiving weight matrices of a layer of a signal generation neural network. Rows of a first matrix for the layer are interleaved by assigning groups of rows of the first matrix to respective thread blocks of a plurality of thread blocks. A first subset of rows of the one or more other weight matrices are assigned to a first subset of the plurality of thread blocks and a second subset of rows of the one or more other weight matrices are assigned to a second subset of the plurality of thread blocks. The first matrix operation is performed substantially in parallel by the plurality of thread blocks. The other matrix operations are performed substantially in parallel by the plurality of thread blocks.

Abstract: A network analysis tool receives network flow information and uses deep learning—machine learning that models high-level abstractions in the network flow information—to identify dependencies between network assets. Based on the identified dependencies, the network analysis tool can discover functional relationships between network assets. For example, a network analysis tool receives network flow information, identifies dependencies between multiple network assets based on evaluation of the network flow information, and outputs results of the identification of the dependencies. When evaluating the network flow information, the network analysis tool can pre-process the network flow information to produce input vectors, use deep learning to extract patterns in the input vectors, and then determine dependencies based on the extracted patterns. The network analysis tool can repeat this process so as to update an assessment of the dependencies between network assets on a near real-time basis.

Abstract: Aspects of the subject technology relate to methods and systems for recommending an energy storage device. In some aspects, a method of the subject technology can include steps for aggregating consumption data for a customer, receiving energy production data for each of a plurality of energy production sources, and determining a utilization efficiency score for each of the plurality of energy storage devices based on the consumption data and the energy production data. In some aspects, methods of the subject technology can also include steps for ranking two or more of the energy storage devices based on the utilization efficiency. In some aspects, machine-readable media are also provided.

Abstract: Provided in the present invention are a method and apparatus for labeling training samples. In the embodiments of the present invention, two mutually independent classifiers, i.e. a first classifier and a second classifier, are used to perform collaborative forecasting on M unlabeled first training samples to obtain some of the labeled first training samples, without the need for the participation of operators; the operation is simple and the accuracy is high, thereby improving the efficiency and reliability of labeling training samples.

Abstract: A method for learning a data embedding network is provided. The method includes steps of: a learning device acquiring and inputting original training data and mark training data into the data embedding network which integrates them and generates marked training data; inputting the marked training data into a learning network which applies a network operation to them and generates 1-st characteristic information, and inputting the original training data into the learning network which applies a network operation to them and generates 2-nd characteristic information; learning the data embedding network such that a data error is minimized, by referring to part of errors referring to the 1-st and the 2-nd characteristic information and errors referring to task specific outputs and their ground truths, and a marked data score is maximized, and learning a discriminator such that a original data score is maximized and the marked data score is minimized.

Abstract: A system for recommending comprises an interface and a processor. An interface is configured to receive an input. The input is stored and associated with two user identifiers. The processor is configured to make recommendations based at least in part on the input.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for modifying a computational graph to include send and receive nodes. Communication between unique devices performing operations of different subgraphs of the computational graph can be handled efficiently by inserting send and receive nodes into each subgraph. When executed, the operations that these send and receive nodes represent may enable pairs of unique devices to conduct communication with each other in a self-sufficient manner. This shifts the burden of coordinating communication away from the backend, which affords the system that processes this computational graph representation the opportunity to perform one or more other processes while devices are executing subgraphs.

Abstract: A second problem Hamiltonian may replace a first problem Hamiltonian during evolution of an analog processor (e.g., quantum processor) during a first iteration in solving a first problem. This may be repeated during a second, or further successive iterations on the first problem, following re-initialization of the analog processor. An analog processor may evolve under a first non-monotonic evolution schedule during a first iteration, and second non-monotonic evolution schedule under second, or additional non-monotonic evolution schedule under even further iterations. A first graph and second graph may each be processed to extract final states versus a plurality of evolution schedules, and a determination made as to whether the first graph is isomorphic with respect to the second graph. An analog processor may evolve by decreasing a temperature of, and a set of quantum fluctuations, within the analog processor until the analog processor reaches a state preferred by a problem Hamiltonian.

Abstract: The present invention relates to a method of associating at least one state in a plurality of states to a new value output by a drifting sensor, the method comprising: /a/ receiving a signal from the sensor, said signal comprising a plurality of values; /b/clustering the values of said signal into a number of clusters equal to the number of the plurality of states, each cluster being associated with a respective state in the plurality of states; /c/ for the new value of the signal, associating at least one state in said plurality of states or a probability rating representing the probability to be associated with one state in said plurality of states for said new value of the signal, the associated state or the associated probability rating being determined based on at least distances (dH, dL) of said new value of the signal to respective clusters.

Abstract: A question answering system that determines whether a question is off-topic by performing the following steps: (i) receiving, by a question answering system, a set of documents; (ii) identifying topical subset(s) for each document of the set of documents using named entity recognition, where each topical subset relates to a corresponding topic; (iii) assigning a set of topic score(s) for each topical subset using natural language processing, where each topic score relates to a corresponding probability associated with the respective topical subset under a probabilistic language model; and (iv) determining, based, at least in part, on the topic score(s) corresponding to the topical subset(s), whether a question input into the question answering system is off-topic.

Abstract: The disclosure herein relates to a quality control method and system in aeronautical manufacturing. The system comprises a tablet connected to a concession management server, the server being itself linked to an aircraft configuration database, a three-dimensional digital model of the aircraft and a fault database. The element to be inspected is identified using the digital model and the fault is characterized by browsing a predetermined decision tree associated with the element.

Abstract: Techniques are described for classifying seasonal patterns in a time series. In an embodiment, a set of time series data is decomposed to generate a noise signal and a dense signal. Based on the noise signal, a first classification is generated for a plurality of seasonal instances within the set of time series data, where each respective instance of the plurality of instances corresponds to a respective sub-period within the season and the first classification associates a first set of one or more instances from the plurality of instances with a particular class of seasonal pattern. Based on the dense signal, a second classification is generated that associates a second set of one or more instances with the particular class. Based on the first classification and the second classification, a third classification is generated, where the third classification associates a third set of one or more instances with the particular class.

Abstract: Method, system, and programs for estimating interests of a plurality of users with respect to a new piece of information are disclosed. In one example, historical interests of the plurality of users are obtained with respect to one or more existing pieces of information. One or more users are selected from the plurality of users. Historical interests of the one or more users can minimize an objective function over the plurality of users. Interests of the one or more users are obtained with respect to the new piece of information. Estimated interests of the plurality of users are generated with respect to the new piece of information based on the obtained interests of the one or more users.

Abstract: The present disclosure describes techniques for object classification using deep forest networks. One example method includes classifying a user object including features associated with the user based on a deep forest network including identifying one or more user static features, one or more user dynamic features, and one or more user association features from the features included in the user object; providing the user static features to first layers, the user dynamic features to second layers, and the user association features to third layers, the first, second, and third layers being different and each providing classification data to the next layer based at least in part on the input data and the provided user features.

Abstract: Techniques are disclosed for improving classification performance in supervised learning. In accordance with some embodiments, a multiclass support vector machine (SVM) having three or more classes may be converted to a plurality of binary problems that then may be reduced via one or more reduced-set methods. The resultant reduced-set (RS) vectors may be combined together in one or more joint lists, along with the original support vectors (SVs) of the different binary classes. Each binary problem may be re-trained using the joint list(s) by applying a reduction factor (RF) parameter to reduce the total quantity of RS vectors. In re-training, different kernel methods can be combined, in accordance with some embodiments. Reduction may be performed until desired classification performance is achieved. The disclosed techniques can be used, for example, to improve classification speed, accuracy, class prioritization, or a combination thereof, in the SVM training phase, in accordance with some embodiments.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for implementing a neural network. In one aspect, the neural network includes a batch renormalization layer between a first neural network layer and a second neural network layer. The first neural network layer generates first layer outputs having multiple components. The batch renormalization layer is configured to, during training of the neural network on a current batch of training examples, obtain respective current moving normalization statistics for each of the multiple components and determine respective affine transform parameters for each of the multiple components from the current moving normalization statistics. The batch renormalization layer receives a respective first layer output for each training example in the current batch and applies the affine transform to each component of a normalized layer output to generate a renormalized layer output for the training example.

Abstract: A method for analyzing and implementing sentiments includes sorting data from the data stream into sorted data by using a corpus builder. The sorted data is then input into an opinion mining platform where selected content is obtained based on the identification of keywords present in the sorted data. A sentiment extraction program generates sentiment metrics based on analysis of the selected content. A rules extractor program determines, based on the sentiment metrics satisfying rules, if actions are to be performed by a business rules engine.

Abstract: Aspects for backpropagation of a convolutional neural network are described herein. The aspects may include a direct memory access unit configured to receive input data from a storage device and a master computation module configured to select one or more portions of the input data based on a predetermined convolution window. Further, the aspects may include one or more slave computation modules respectively configured to convolute one of the one or more portions of the input data with one of one or more previously calculated first data gradients to generate a kernel gradient, wherein the master computation module is further configured to update a prestored convolution kernel based on the kernel gradient.