Abstract: A system and method simulates conversation with a human user. The system and method receive media, convert the media into a system-specific format, and compare the converted media to a vocabulary. The system and method generate a plurality of intents and a plurality of sub-entities and transform them into a pre-defined format. The system and method route intents and the sub-entities to a first selected knowledge engine and a second knowledge engine. The first selected knowledge engine selects the second knowledge engine and each active grammar in the vocabulary uniquely identifies each of the knowledge engines.

Abstract: A model optimization system can reduce the delays caused by cache misses and page faults by converting a model of one or more decision trees into machine code that is optimized to avoid these memory faults. The model optimization system can convert a model into machine code by converting each tree of the model into a series of nested if/then statements and converting each series of nested if/then statements into optimized machine code. In some implementations, the model can be converted into optimized machine code only when an amount of processing required to convert the model into the optimized machine code is less than the expected cost savings of using the optimized machine code, instead of an unmodified version of the model, over the life of the model.

Abstract: A spiking neural network device including a spiking neural network circuit including a crossbar array of plural synapses; plural axons connected with the spiking neural network circuit, the plural axons receiving input of a spike signal; and plural Poisson spike generators respectively provided for the plural axons. Each Poisson spike generator can be set whether or not to emit the spike signal based on an input signal to be processed, and each Poisson spike generator can, be set to emit the spike signal being configured to generate a Poisson spike train different from each other. and supply the Poisson spike train to a corresponding one of the plural axons.

Abstract: In one embodiment, a device groups feature vectors representing network traffic flows into bags. The device forms a bag representation of a particular one of the bags by aggregating the feature vectors in the particular bag. The device extends one or more feature vectors in the particular bag with the bag representation. The extended one or more feature vectors are positive examples of a classification label for the network traffic. The device trains a network traffic classifier using training data that comprises the one or more feature vectors extended with the bag representation.

Abstract: A novel system and method are described that allows for implementation of compact and efficient deep learning AI solutions to advanced sensor signal processing functions. The process includes the following stages: (1) A method for generating requisite annotated training data in sufficient quantity to ensure convergence of a deep learning neural network (DNN); (2) A method for implementing the resulting DNN onto a Spiking Neural Network (SNN) architecture amenable to efficient neuromorphic integrated circuit (IC) architectures; (3) A method for implementing the solution onto a neuromorphic IC; and (4) A statistical method for ensuring reliable performance.

Abstract: Systems and methods are provided for embedded data inference. The systems and methods may process camera and other sensor data in by leveraging processing and storage capacity of one or more devices nearby or in the cloud to augment or update the sensor processing of an embedded device. The joint processing may be used in stationary cameras or in vehicular systems such as cars and drones, and may improve crop assessments, navigation, and safety.

Abstract: Methods, systems and computer program products for resolving multiple magnitudes assigned to a target vector are disclosed. A target vector that includes one or more target vector dimensions is received. One of the target vector dimensions is processed to determine a total number of magnitudes assigned to the processed target vector dimension. Also, a source vector that includes one or more source vector dimensions is received. The received source vector is processed to determine a total number of features associated with the source vector. When it is detected that the total number of magnitudes assigned to the processed target vector dimension exceeds one, one of the assigned magnitudes is selected based on one of the determined features associated with the source vector.

Abstract: Systems, apparatuses, and methods for implementing memory bandwidth reduction techniques for low power convolutional neural network inference applications are disclosed. A system includes at least a processing unit and an external memory coupled to the processing unit. The system detects a request to perform a convolution operation on input data from a plurality of channels. Responsive to detecting the request, the system partitions the input data from the plurality of channels into 3D blocks so as to minimize the external memory bandwidth utilization for the convolution operation being performed. Next, the system loads a selected 3D block from external memory into internal memory and then generates convolution output data for the selected 3D block for one or more features. Then, for each feature, the system adds convolution output data together across channels prior to writing the convolution output data to the external memory.

Abstract: An energy management system includes a neural network, a predictive model, and a dictionary reducer. The network iteratively calculates weights, resulting in a final set, for each of a plurality of single-word terms taken from training data business names, where each of the weights is indicative of a likelihood of correlating one of a plurality of business categories. The predictive employs sets of the weights to predict a first corresponding one of the plurality of business categories for each of the training data business names until employment of the final set accurately predicts a correct business category for the each of the training data business names, and subsequently employs the final set of the weights to predict a second corresponding one of the plurality of business categories for each of a plurality of operational business names. The dictionary reducer eliminates unessential terms taken to determine the plurality of single-word terms.

Abstract: Embodiments of present disclosure discloses system and method for managing applications in electronic device. Initially, relation tree associated with applications is identified. The relation tree is generated based on learning technique implemented for applications, parameters, enablers associated with electronic device. Based on identified relation tree, enablers are identified from plurality of enablers, corresponding to each of applications. Further, current status of parameters based on current status of enablers is retrieved. The relation tree is updated based on learning technique implemented for at least one of the current status of the parameters, new applications, new parameters, enablers associated with the electronic device. An application from the applications is identified based on the current status of the parameters and the relation tree. The electronic device is instructed to perform operations associated with the identified application.

Abstract: Hardware placement of neural networks is provided. In various embodiments, a network description is read. The network description describes a spiking neural network. The neural network is trained. An initial placement of the neural network on a plurality of cores is performed. The cores are located on a plurality of chips. Inter-chip communications are measured based on the initial placement. A final placement of the neural network on the plurality of cores is performed based on the inter-chip communications measurements and the initial placement. The final placement reduces inter-chip communication.

Abstract: Methods and apparatus are provided for implementing propagation of probability distributions of random variables over a factor graph. Such a method includes providing a spiking neural network, having variable nodes interconnected with factor nodes, corresponding to the factor graph. Each of the nodes comprises a set of neurons configured to implement computational functionality of that node. The method further comprises generating, for each of a set of the random variables, at least one spike signal in which the probability of a possible value of that variable is encoded via the occurrence of spikes in the spike signal, and supplying the spike signals for the set of random variables as inputs to the neural network at respective variable nodes. The probability distributions are propagated via the occurrence of spikes in signals propagated through the neural network.

Abstract: A computer-implemented method according to one embodiment includes identifying a first classifier training data element and a second classifier training data element, calculating a similarity metric between the first classifier training data element and the second classifier training data element, and determining a classification for the first classifier training data element and the second classifier training data element, utilizing the similarity metric between the first classifier training data element and the second classifier training data element.

Abstract: Techniques that facilitate improving an efficiency of a neural network are described. In one embodiment, a system is provided that comprises a memory that stores computer-executable components and a processor that executes computer-executable components stored in the memory. In one implementation, the computer-executable components comprise an initialization component that selects an initial value of an output limit, wherein the output limit indicates a range for an output of an activation function of a neural network. The computer-executable components further comprise a training component that modifies the initial value of the output limit during training to a second value of the output limit, the second value of the output limit being provided as a parameter to the activation function. The computer-executable components further comprise an activation function component that determines the output of the activation function based on the second value of the output limit as the parameter.

Abstract: Embodiment of the invention are directed to transmitting signals between neurons of a hardware-implemented, spiking neural network (or SNN). The network includes neuronal connections, each including a synaptic unit connecting a pre-synaptic neuron to a post-synaptic neuron. Spikes received from the pre-synaptic neuron of said each neuronal connection are first modulated, in frequency, based on a synaptic weight stored on said each synaptic unit, to generate post-synaptic spikes, such that a first number of spikes received from the pre-synaptic neuron are translated into a second number of post-synaptic spikes. At least some of the spikes received from the pre-synaptic neuron may, each, be translated into a train of two or more post-synaptic spikes. The post-synaptic spikes generated are subsequently transmitted to the post-synaptic neuron of said each neuronal connection. The novel approach makes it possible to obtain a higher dynamic range in the synapse output.

Abstract: Systems and methods are provided for detecting events in industrial processes. An acquisition system may include one of a camera and an audio recorder to acquire monitoring data in the form of one of imaging data and acoustic data, respectively. A computer system, may include a machine learning engine and may be programmed to classify the monitoring data under a classifier, quantify, based on the classifier, the monitoring data with at least one quantifier, and detect an event when the at least one quantifier satisfies a predetermined rule corresponding to the at least one quantifier.

Abstract: A method and a cloud-computing architecture for enabling dynamic access of an artificial intelligence engine are described. A record that includes a set of one or more fields is stored in a database. A first field from the set of fields includes an identification of an artificial intelligence (AI) engine and one or more additional fields from the set of fields respectively include one or more parameters for the AI engine. The record is accesses causing the AI engine to run with the one or more parameters. As a result of the AI engine running with the one or more parameters upon access of the record, a desired predicted output is obtained.

Abstract: Identification of content classes deemed important by users is received. Classifiers are trained on user communication text to recognize content matching content classes. A composite classifier is applied to text of communication among users of a second set of users, including tagging, by the composite classifier, content in text of communication among the users of the second set. The tagging is responsive to the classifiers recognizing their respective content classes in text of communication among the users of the second set.

Abstract: In one example, an integrated circuit includes a first circuit, a second circuit, a third circuit, and a fourth circuit. The first circuit is configured to receive an input value and generate a first intermediate value based on a first probability density distribution associated with the input value. The second circuit comprises a set of multiplexer circuits configured to select, from a first set of candidate values and based on the first intermediate value, a first product of the first intermediate value and a weight value. The third circuit is configured to generate a second intermediate value based on a sum of the first product and a second product received from another circuit. The fourth circuit is configured to generate an output value based on the second intermediate value and a second probability density distribution associated with the second intermediate value.

Abstract: A computer-implemented method for dual sequence inference using a neural network model includes generating a codependent representation based on a first input representation of a first sequence and a second input representation of a second sequence using an encoder of the neural network model and generating an inference based on the codependent representation using a decoder of the neural network model. The neural network model includes a plurality of model parameters learned according to a machine learning process. The encoder includes a plurality of coattention layers arranged sequentially, each coattention layer being configured to receive a pair of layer input representations and generate one or more summary representations, and an output layer configured to receive the one or more summary representations from a last layer among the plurality of coattention layers and generate the codependent representation.