Abstract: Features are disclosed for reducing the dynamic range of an approximated trained artificial neural network weight matrix in an automatic speech recognition system. The weight matrix may be approximated as two low-rank matrices using a decomposition technique. This approximation technique may insert an additional layer between the two original layers connected by the weight matrix. The dynamic range of the low-rank decomposition may be reduced by applying the square root of singular values, combining them with both low-rank matrices, and utilizing a random rotation matrix to further compress the low-rank matrices. Reduction of dynamic range may make fixed point scoring more effective due to smaller quantization error, as well as make the neural network system more favorable for retraining after approximating a neural network weight matrix. Features are also disclosed for adjusting the learning rate during retraining to account for the low-rank approximations.

Abstract: A pattern based audio searching method includes labeling a plurality of source audio data based on patterns to obtain audio label sequences of the source audio data; obtaining, with a processing device, an audio label sequence of target audio data; determining matching degree between the target audio data and the source audio data according to a predetermined matching rule based on the audio label sequence of the target audio data and the audio label sequences of the source audio data; and outputting source audio data having matching degree higher than a predetermined matching threshold as a search result.

Abstract: A neural network comprises a plurality of artificial neurons and a plurality of artificial synapses each input neuron being connected to each output neuron by way of an artificial synapse, the network being characterized in that each synapse consists of a first memristive device connected to a first input of an output neuron and of a second memristive device, mounted in opposition to said first device and connected to a second, complemented, input of said output neuron so that said output neuron integrates the difference between the currents originating from the first and second devices.

Abstract: Systems and methods for determining predictive model types are provided. A method may include generating a predictive model for a web page of a website, wherein the web page includes a configuration defining one or more objects presented with the web page, and wherein each object is associated with a predictive model. The method may include determining one or more predictive model types that are associated with the predictive model, determining one or more performance indicators that correspond to each determined predictive model type, wherein performance indicators represent one or more benefits to a website, selecting a predictive model type of the predictive model out of the one or more predictive model types, wherein the predictive model type is selected based on a performance indicator corresponding to the selected predictive model type, and determining a configuration of the web page using the selected predictive model type of the predictive model.

Abstract: An affordable artificial intelligent (AI) computer is invented by combining present computer with a parallel hardware search system. Such a computer can be treated as a Turing Machine. Instead of sequentially processing computer instructions, this computer executes AI logic reasoning. The parallel hardware search system use pure parallel hardware to execute virtual B-tree search. Hierarchical page table and hash techniques are also used for very large data volume. The prototype of this invented system is successfully built into a PCIE card which mainly contains a Xilinx's Kintex7 FPGA chip and two DDR3 memory modules. FPGA chip includes: 32 32-bit processing units (PUs), one PCIE controller, one search/delete/insert controller and two DDR3 controllers.

Abstract: Semantic indexing methods and systems are disclosed. One such method is directed to training a semantic indexing model by employing an expanded query. The query can be expanded by merging the query with documents that are relevant to the query for purposes of compensating for a lack of training data. In accordance with another exemplary aspect, time difference features can be incorporated into a semantic indexing model to account for changes in query distributions over time.

Abstract: In some embodiments, techniques for rationalizing a recommendation include determining a recommended item for a user using a first recommendation engine, wherein the first recommendation engine receives as an input first behavioral data associated with a user, and generates an identifier corresponding to a recommended item; creating a rationalization for the recommended item using a first rationalization engine, wherein the first rationalization engine receives as inputs the recommended item and second behavioral data associated with the user, and generates a rationalization, wherein the rationalization includes a constructed rationalization of why the recommended item is recommended for the user, and wherein the creation of the rationalization is not based solely on an actual reason the first recommendation engine determined the recommended item; associating the rationalization for the recommended item with the recommended item; and providing the recommended item and associated rationalization.

Abstract: Online learning of an ensemble of classifiers or regressors is performed to predict a non-stationary time-varying parameter over a time series of episodes. For an episode, an ensemble action is selected from a set of ensemble actions based on ensemble action quality values (Q values) at an error state for the episode. The selected ensemble action is executed to update the ensemble. A cost of executing the selected ensemble action is computed or retrieved, and a reward is computed indicating how well the updated ensemble predicts the non-stationary time-varying parameter over the episode. The Q value for the selected ensemble action at the error state for the episode is updated based on both the reward and the cost of executing the selected ensemble action. The cost may be based on the ensemble action alone, or on both the ensemble action and the error state for the episode.

Abstract: This invention relates to a method and apparatus for characterizing composite materials, and in particular, to utilizing an artificial neural network for predicting an impact resistance of a composite material. A method for predicting an impact resistance of a composite material in accordance with the present invention includes the steps of designing an artificial neural network including a plurality of neurons, training the artificial neural network to predict the impact resistance by adjusting an output of the plurality of neurons according to sample data and known results of the sample data, inputting data of the composite material into the artificial neural network, and utilizing the artificial neural network to predict the impact resistance of the composite material.

Abstract: Embodiments of the invention provide a neural core circuit comprising a synaptic interconnect network including plural electronic synapses for interconnecting one or more source electronic neurons with one or more target electronic neurons. The interconnect network further includes multiple axon paths and multiple dendrite paths. Each synapse is at a cross-point junction of the interconnect network between a dendrite path and an axon path. The core circuit further comprises a routing module maintaining routing information. The routing module routes output from a source electronic neuron to one or more selected axon paths. Each synapse provides a configurable level of signal conduction from an axon path of a source electronic neuron to a dendrite path of a target electronic neuron.

Abstract: There is provided a frequent pattern extraction apparatus. In the apparatus, time series data for an item operation is divided into a plurality of item operation sets. Using a set degree of similarity between the item operation sets, an abstract item operation set is generated. A pattern of sequences that frequently appear is extracted from the sequences in the abstract item operation set. Using the pattern of sequences that frequently appear, an item as well as an item operation are recommended to a user.

Abstract: A reasoning engine is disclosed. Contemplated reasoning engines acquire data relating to one or more aspects of various environments. Inference engines within the reasoning engines review the acquire data, historical or current, to generate one or more hypotheses about how the aspects of the environments might be correlated, if at all. The reasoning engine can attempt to validate the hypotheses through controlling acquisition of the environment data.

Abstract: A monitoring system for determining the future operational condition of an object includes an empirical model to receive reference data that indicates the normal operational state of the object and input pattern arrays. Each input pattern array has a plurality of input vectors, while each input vector represents a time point and has input values representing a plurality of parameters indicating the current condition of the object. The model generates estimate values based on a calculation that uses an input pattern array and the reference data to determine a similarity measure between the input values and reference data. The estimate values, in the form of an estimate matrix, include at least one estimate vector of inferred estimate values, and represents at least one time point that is not represented by the input vectors. The inferred estimate values are used to determine a future condition of the object.

Abstract: Embodiments of the invention provide a neural network comprising multiple functional neural core circuits, and a dynamically reconfigurable switch interconnect between the functional neural core circuits. The interconnect comprises multiple connectivity neural core circuits. Each functional neural core circuit comprises a first and a second core module. Each core module comprises a plurality of electronic neurons, a plurality of incoming electronic axons, and multiple electronic synapses interconnecting the incoming axons to the neurons. Each neuron has a corresponding outgoing electronic axon. In one embodiment, zero or more sets of connectivity neural core circuits interconnect outgoing axons in a functional neural core circuit to incoming axons in the same functional neural core circuit.

Abstract: A monitoring system for determining the future behavior of a financial system includes an empirical model to receive reference data that indicates the normal behavior of the system and input pattern arrays. Each input pattern array has a plurality of input vectors, while each input vector represents a time point and has input values representing a plurality of parameters indicating the current condition of the system. The model generates estimate values based on a calculation that uses an input pattern array and the reference data to determine a similarity measure between the input values and reference data. The estimate values, in the form of an estimate matrix, include at least one estimate vector of inferred estimate values, and represents at least one time point that is not represented by the input vectors. The inferred estimate values are used to determine a future behavior of the financial system.

Abstract: A method for recommending an application includes obtaining an input model representing user interaction patterns during execution of a first application. The input model is compared to a reference model representing user interaction patterns during execution of a second application. A similarity is determined between the input model and the reference model. A recommendation of the second application is generated in response to the similarity.

Abstract: In some embodiments, techniques for rationalizing a recommendation include determining a recommended item for a user using a first recommendation engine, wherein the first recommendation engine receives as an input first behavioral data associated with a user, and generates an identifier corresponding to a recommended item; creating a rationalization for the recommended item using a first rationalization engine, separate from the recommendation engine, wherein the first rationalization engine receives as inputs the recommended item and second behavioral data associated with the user, and generates a rationalization, wherein the rationalization includes a constructed rationalization of why the recommended item is recommended for the user, and wherein the creation of the rationalization is not based on an actual reason the first recommendation engine determined the recommended item; associating the rationalization for the recommended item with the recommended item; and providing the recommended item and associa

Abstract: An improved technique generates questions to authenticate a user as part of a group. Along these lines, a KBA system, upon receiving a request to authenticate a particular user, collects facts having references to users of the group of users. The collected facts, however, may also include references to users not in the group of users. In building a set of questions for the particular user, the KBA system is capable of favoring facts having references to users of the group of users and few, if any, references to users not in the group of users; conversely, the KBA system is capable of discarding facts having too many references to users not in the group of users. The particular user's responses to the set of questions are indicative of whether the particular user belongs to the group.

Abstract: Methods, systems, and apparatus for computing quality scores based on neighbor features. In one aspect, a method includes obtaining a quality model that was trained using a set of training entities; identifying a set of candidate entities that are different from each of the training entities; for each candidate entity: obtaining a first quality score for the candidate entity; obtaining one or more neighbor features for neighbor entities of the candidate entity, where each neighbor entity of the candidate entity is linked to the candidate entity; obtaining one or more entity specific feature values for the candidate entity, where each entity specific feature value is determined independent of the neighbor entities of the candidate entity; and determining a second quality score for the candidate entity using the quality model, the second quality score being computed based on the first quality score, the neighbor features, and the entity specific feature values.

Abstract: A system and method for providing multi-dimensional context-aware adaptation in vehicular networks is disclosed. The system comprises a collection module, a context resolving module, a parameter determination module and a distribution module. The collection module collects context data describing a context in a communication environment. The context resolving module resolves the context data to a matching historical context and determines one or more historical context groups associated with the matching historical context. The parameter determination module determines a subset of operating parameters from the one or more historical context groups. The distribution module distributes the subset of operating parameters to a network stack communication module.