Simple principles of cognitive computation with distributed representations
Sammanfattning: Brains and computers represent and process sensory information in different ways. Bridgingthat gap is essential for managing and exploiting the deluge of unprocessed andcomplex data in modern information systems. The development of brain-like computersthat learn from experience and process information in a non-numeric cognitive way willopen up new possibilities in the design and operation of both sensor and informationcommunication systems.This thesis presents a set of simple computational principles with cognitive qualities,which can enable computers to learn interesting relationships in large amounts of datastreaming from complex and changing real-world environments. More specifically, thiswork focuses on the construction of a computational model for analogical mapping andthe development of a method for semantic analysis with high-dimensional arrays.A key function of cognitive systems is the ability to make analogies. A computationalmodel of analogical mapping that learns to generalize from experience is presented in thisthesis. This model is based on high-dimensional random distributed representations anda sparse distributed associative memory. The model has a one-shot learning process andan ability to recall distinct mappings. After learning a few similar mapping examplesthe model generalizes and performs analogical mapping of novel inputs. As a majorimprovement over related models, the proposed model uses associative memory to learnmultiple analogical mappings in a coherent way.Random Indexing (RI) is a brain-inspired dimension reduction method that was developedfor natural language processing to identify semantic relationships in text. Ageneralized mathematical formulation of RI is presented, which enables N-way RandomIndexing (NRI) of multidimensional arrays. NRI is an approximate, incremental, scalable,and lightweight dimension reduction method for large non-sparse arrays. In addition, itprovides low and predictable storage requirements, and also enables the range of arrayindices to be further extended without modification of the data representation. Numericalsimulations of two-way and ordinary one-way RI are presented that illustrate whenthe approach is feasible. In conclusion, it is suggested that NRI can be used as a tool tomanage and exploit Big Data, for instance in data mining, information retrieval, socialnetwork analysis, and other machine learning applications.
KLICKA HÄR FÖR ATT SE AVHANDLINGEN I FULLTEXT. (PDF-format)