Introduction to Cognitive ScienceandAdaptive Resonance Theory (ART) :Introduction to Cognitive ScienceandAdaptive Resonance Theory (ART) Mete Balcı Presented at Advanced Artificial Intelligence Class, Fall 2005-2006, İTÜ.


Agenda :Agenda Introduction to Cognitive Science Definitions Nerve Cell and Brain Anatomy Problem Adaptive Resonance Theory (ART) Ideas behind ART What is ART ? Details Application A Model to Solve “Tower of Hanoi” as a Behavioural Task


Cognitive Science :Cognitive Science Definition: Cognitive (in Oxford English Dictionary, 1586): “to the action or process of knowing” Scientific study either of “Mind” or “Intelligence” (Luger, 1994).


Cognitive Science :Cognitive Science Definition: Cognitive (in Oxford English Dictionary, 1586): “to the action or process of knowing” Scientific study either of “Mind” or “Intelligence” (Luger, 1994). Highly Interdisciplinary Psychology, Neuroscience, Biology, Philosophy, Computer Science, Linguistics, Robotics, Mathematics... The shared interest that has produced this coalition is understanding the nature of the mind.


Cognitive Science :Cognitive Science Definition: Cognitive (in Oxford English Dictionary, 1586): “to the action or process of knowing” Scientific study either of “Mind” or “Intelligence” (Luger, 1994). Highly Interdisciplinary Psychology, Neuroscience, Biology, Philosophy, Computer Science, Linguistics, Robotics, Mathematics... The shared interest that has produced this coalition is understanding the nature of the mind. Approaches Symbolic Connectionist Dynamic Systems


Cognitive Science :Cognitive Science Levels Behavioral [Computational] : What is the output ? Functional [Algorithmic] : How is the output produced ? Physical [Implementational] : What produces the output ?


Cognitive Science :Cognitive Science Levels Behavioral [Computational] : What is the output ? Functional [Algorithmic] : How is the output produced ? Physical [Implementational] : What produces the output ? Research Methods Behavioural Experiments Reaction Times, Accuracy Brain Imaging EEG, CT, MRI, fMRI, PET Computational Modelling Neurobiological Methods Single Cell Recordings, Animal Models, Lesion Patients


AI :AI Involves: The study of cognitive phenomena in machines. Deals with: Intelligent behavior, learning and adaptation in machines. Approaches Symbolic Machine learning, Statistical analysis Expert Systems Bayesian Networks Non-symbolic Iterative development or learning, Soft computing Neural Networks Genetic Algorithms The field of cognitive science overlaps AI.


Nerve Cell Anatomy :Nerve Cell Anatomy


Brain Anatomy :Brain Anatomy


Brain Anatomy :Brain Anatomy


Brain Anatomy :Brain Anatomy Ungerleider & Mishkin, 1982.


MacaqueBrain :MacaqueBrain Felleman, D. J. and Van Essen, D. C. Cerebral Cortex 1:1-47. 1991


Problem :Brain / Mind Problem


Problem :Brain / Mind How does the brain control behavior ? Problem


Problem :Brain / Mind How does the brain control behavior ? How can technology emulate biological intelligence ? Problem


Stephen Grossberg :Stephen Grossberg Founder of ART Chairman, Department of Cognitive and Neural Systems Boston University http://cns.bu.edu/Profiles/Grossberg/


According to Grossberg :According to Grossberg Fundamental problem of perception and cognition: How humans discover, learn and recognize invariant properties of the environments to which they are exposed ?


According to Grossberg :According to Grossberg Fundamental problem of perception and cognition: How humans discover, learn and recognize invariant properties of the environments to which they are exposed ? In AI, we are trying to find methods to extract invariant properties of subject we are working on.


According to Grossberg :According to Grossberg Fundamental problem of perception and cognition: How humans discover, learn and recognize invariant properties of the environments to which they are exposed ? How do we continue to learn throughout life ?


According to Grossberg :According to Grossberg Fundamental problem of perception and cognition: How humans discover, learn and recognize invariant properties of the environments to which they are exposed ? How do we continue to learn throughout life ? Is it possible to design a system that learns continously without forgetting the past ?


According to Grossberg :According to Grossberg When such recognition codes spontaneously emerge through an individual’s interaction with an environment, the processes are said to undergo “SELF-ORGANIZATION”.


According to Grossberg :According to Grossberg When such recognition codes spontaneously emerge through an individual’s interaction with an environment, the processes are said to undergo “SELF-ORGANIZATION”. ART is a type of neural network developed by a theory of how recognition codes are self-organized.


According to Grossberg :According to Grossberg The network model is capable of self-organizing, self-stabilizing, and self-scaling its recognition codes in response to arbitrary temporal sequences of arbitrarily many input patterns of variable complexity.


Stability-Plasticity Dilemma :Stability-Plasticity Dilemma Plasticity: Each system generates recognition codes that are new steady states and basins of attractions corresponding to the critical feature patterns, or prototypes, that represent invariants of the set of all experienced input patterns.


Stability-Plasticity Dilemma :Stability-Plasticity Dilemma Plasticity: Each system generates recognition codes that are new steady states and basins of attractions corresponding to the critical feature patterns, or prototypes, that represent invariants of the set of all experienced input patterns. Plasticity is required for learning !


Stability-Plasticity Dilemma :Stability-Plasticity Dilemma Plasticity: Each system generates recognition codes that are new steady states and basins of attractions corresponding to the critical feature patterns, or prototypes, that represent invariants of the set of all experienced input patterns. Stability: The learned codes are buffered against irrelevant inputs. The formation of steady states is internally controlled using mechanisms that suppress possible sources of system instability.


Stability-Plasticity Dilemma :Stability-Plasticity Dilemma Plasticity: Each system generates recognition codes that are new steady states and basins of attractions corresponding to the critical feature patterns, or prototypes, that represent invariants of the set of all experienced input patterns. Stability: The learned codes are buffered against irrelevant inputs. The formation of steady states is internally controlled using mechanisms that suppress possible sources of system instability. Stability is required for surviving !


What is ART ? :What is ART ? Adaptive Resonance Theory (ART) Proposed by S. Grossberg in late 60s. First complete article in 1976. More than 200 articles in 40 years. Carpenter, G.A. & Cohen, M.A.


What is ART ? :What is ART ? Adaptive Resonance Theory (ART) Based on Classical Conditioning Paradigms Solid Mathematical Theory Capable of working in real time with real world data Basic theory is extended to explain different phenomena.


ART Details :ART Details Input Stimuli


ART Details :ART Details Input Stimuli


ART Details :ART Details Input Stimuli


ART Details :ART Details Input Stimuli


ART Details :ART Details Input Stimuli


ART Details :ART Details Input Stimuli


ART Details :ART Details Input Stimuli


ART Details :ART Details Input Stimuli Attention Subsystem Orienting Subsystem


ART Details :ART Details Attention Subsystem Orienting Subsystem Short Term Memory Short Term Memory Long Term Memory Control Unit Control Unit F1 F2 Input Stimuli


ART Details :ART Details


ART Details :ART Details


ART Details :ART Details


ART Details :ART Details


ART Details :ART Details


ART Details :ART Details


ART Details :ART Details


ART Details :ART Details


ART Details :ART Details


ART Details :ART Details


ART Details :ART Details


ART Details :ART Details


ART Details :ART Details


ART as a Dynamic System :ART as a Dynamic System I


ART :ART Cognitive Science Neocortex Hippocampus LGN (Lateral Geniculate Nucleus) Visual Cortex Engineering Pattern Recognition/Classification


Contemporary ART Studies :Contemporary ART Studies ARTMAP, 1991 Supervised Learning and Classification ARTSTREAM, 2004 Auditory Stream Analysis Laminar Computing How laminar frontal cortex and basal ganglia circuits interact to control planned and reactive saccades, Neural Networks’ 04.


Contemporary ART Studies :Contemporary ART Studies ARTMAP, 1991 Supervised Learning and Classification ARTSTREAM, 2004 Auditory Stream Analysis Laminar Computing How laminar frontal cortex and basal ganglia circuits interact to control planned and reactive saccades, Neural Networks’ 04.


Tower of Hanoi (ToH) :Tower of Hanoi (ToH) The ToH is a behaviourally rich planning task that is used to assess the planning ability of the prefrontal cortex (PFC).


Model to Solve ToH :Model to Solve ToH Balci, M. & Sengor, N. A Connectionist Model of Planning: A Behavioural Approach. Computational Cognitive Neuroscience, November 10-11, 2005. Washington DC, USA.


Problems :Problems


Simulation Results :Simulation Results


References :References Grossberg, S., Adaptive Pattern Classification and Universal Recoding: I-II. Biological Cybernetics. 1976. Grossberg, S., How does a brain build a cognitive code ?. Psychological Review. 1980. Carpenter, G.A. & Grossberg, S., A Massively Parallel Architecture for a Self-Organizing Neural Pattern Recognition Machine. Computer, Vision, Graphics, and Image Processing. 1987. Grossberg, S., The Link Between Brain, Learning, Attention, and Consciousness. Boston University Technical Report CAS/CNS-TR-97-018. 1998.


Thank you ! :Thank you ! Mete Balcı mete@ieee.org http://www.metebalci.com/pubs Questions ?