How to build the cognitive network?

Abstract

Event Abstract Back to Event How to build the cognitive network? Plewczynski Dariusz1* 1 University of Warsaw, Interdisciplinary Centre for Mathematical and Comp, Poland How to build the cognitive network? What is the influence of the learning network on the final quality of information processing in cognitive systems? How a node of such network has to be designed in order to effectively process incoming information that is both time dependent and spatially heterogenous? We address those questions by benchmarking different machine learning algorithms using real life examples (mostly from bioinformatics and chemoinformatics) that mimics the different pattern recognition tasks typically used in various machine learning applications. In the framework of Agent Based Modelling (ABM) the most promising cognitive network appears to be constructed from an sparse ensemble of interacting agents. The cognitive networks are able to successfully combine outcomes of individual learning agents in order to determine the final decision with the higher accuracy than any single learning method used in the construction of a network. We discuss here also the mean-field approximation of the cognitive networks in the limit of the infinite number of learning agents. The ensemble of interacting learning agents, acquires and process incoming information using various types, or different versions of machine learning and clustering algorithms. The abstract learning space, where all agents are located, is constructed here using a randomly connected, sparse model. It couples locally the certain small percentage of similar learning agents, yet also connect remotely about 10% of the other agents with random strength values. Such network simulates the higher level integration of information acquired from the independent learning trials. The final classification of incoming input data is therefore defined as the stationary state of the cognitive system using simple majority rule, yet the minority clusters that share opposite classification outcome can be observed in the system. The cognitive network is able to couple different scales of both space and time patterns by assigning them to different subsets of learning nodes. The probability of selecting proper class for a given input data, can be estimated even without the prior knowledge of its affiliation. The fuzzy logic can be easily introduced into the system, even if learning agents are build from simple binary classification machine learning algorithms by calculating the percentage of agreeing agents. The model is inspired by the large scale cognitive computing models of the brain cortex areas (D. Modha et al., Izhikevich et al.).