Abstract: Method and system of training a student neural network (SNN) model. A first training phase is performed over a plurality of epochs during which a smoothing factor to teacher neural network (TNN) model outputs to generate smoothed TNN model outputs, a first loss is computed based on the SNN model outputs and the smoothed TNN model outputs, and an updated set of the SNN model parameters is computed with an objective of reducing the first loss in a following epoch of the first training phase. The soothing factor is adjusted over the plurality of epochs of the first training phase to reduce a smoothing effect on the generated smoothed TNN model outputs. A second training phase is performed based on the SNN model outputs and a set of predefined expected outputs for the plurality of input data samples.

Abstract: The system and method of present invention provides natural language processing of queries associated with one or more domains via individually configurable decentralized AI-based scalable units leading to flexibility in scaling and descaling. In operation, the present invention provides for hosting NLP resources configurable to perform natural language processing of queries and generate responses associated with respective domain. Each NLP resource comprises logically independent AI-based units. Each AI-based unit is configurable to perform processing of queries and emulate responses associated with respective subdomain of the domain of corresponding NLP resource. Further, each AI-based unit comprises respective one or more AI-based micro-units configurable to perform processing of queries and emulate responses associated with respective one or more topics of the subdomains of corresponding AI-based unit.

Abstract: In one embodiment, a search engine may generate and store a plurality of search index segments such that each of the search index segments is stored in a corresponding one of a plurality of heaps of memory. The plurality of search index segments may include inverted index segments mapping content to documents containing the content. A garbage collection module may release one or more heaps of the memory.

Abstract: In one embodiment, a search engine may generate and store a plurality of search index segments such that each of the search index segments is stored in a corresponding one of a plurality of heaps of memory. The plurality of search index segments may include inverted index segments mapping content to documents containing the content. A garbage collection module may release one or more heaps of the memory.

Abstract: A peer node configured to exchange data with other peer nodes through a peer-to-peer protocol includes a categorization and selection module. The selection module is configured to receive peer node identifiers extracted from messages received from the other peer nodes, to categorize the peer node identifiers according to one or more criteria in one of at least two categories, and to select a peer node identifier as a peer node identifier to be used for communication through the peer-to-peer protocol, if the peer node identifier was categorized in a first category of the at least two categories.

Abstract: Processing method for replying to a distributed hash table (DHT) message from a DHT client in order to allow the DHT client to fill his DHT routing table, includes at a proxy processing agent having a plurality of nodes, receiving from the DHT client a DHT request message, and generating at least one node identifier and associating each node identifier with an IP address and a port number. Each IP address/port number combination corresponding with a node of the plurality of nodes. A DHT response message containing the at least one node identifier and associated IP address and port is sent to the DHT client, and the at least one generated node identifier and associated IP address and port are stored in connection with said DHT client.