Stimulus Processing in Autonomously Active Cognitive Systems - - PowerPoint PPT Presentation

Stimulus Processing in Autonomously Active Cognitive Systems

Claudius Gros

Institute for Theoretical Physics J.W. Goethe University Frankfurt

» the cognitive system approach to AI – an overview«

1

pitfalls of traditional AI

‘mainstream AI will not lead to human-level cognitive systems (CS)’

• the architectural conundrum

AI: algorithmic optimization CS: cognitive capabilities emergent from universal principles

• the motivational problem

AI: tasks given by external supervisor CS: diffusive emotional control

2

the architectural conundrum

cognitive systems - universal principles

• universal time prediction tasks (Elman)

⊲ environmental model building ⊲ unsupervised generation of abstract concepts

• behavioral complexity optimization (Edelman, Sporns, ...)

⊲ spontaneous explorative strategies

• autonomous internal dynamics (Gros)

⊲ semantic learning: unsupervised, developmental

.... are all applicable for a wide range of environments

• C. Gros

“Cognitive computation with autonomously active neural networks: An emerging field”, Cognitive Computation 1, 77 (2009)

3

the motivational problem

cognitive systems as living & embedded dynamical systems

• proprioceptual survival parameters

⊲ blood pressure, blood sugar level, pain signals, ... ⊲ ‘survival instinct’

• diffusive emotional control - neuromodulators

⊲ signaling: novelty, learning, ... ⊲ diffusive: acts on entire neural ensembles ⊲ meta-learning: thresholds, synaptic plasticities, ... ⊲ evolved from neutral homeostatic regulation

• C. Gros

“Emotions, diffusive emotional control and the motivational problem for autonomous cognitive systems”, Handbook of Research on Synthetic Emotions and Sociable Robotics: New Applications in Affective Computing and Artificial Intelligence, Vallverdí, Casacuberta (Eds.) (2009, in print)

4

information processing vs. diffusive control

module A module B module C module D cognitive processing cognitive processing cognitive processing information flow diffusive control diffusive signaling diffusive control status signals diffusive control emotional interactions diffusive control cognitive processing

Frankfurt Cognitive System Platform

⊲ full online architectural

configurability

⊲ learning exclusively

: unsupervised : online current status

⊲ autonomous dynamics

: transient states : stimulus processing : novelty signals » emergent cognitive capability: non-linear ICA «

5

correlations: internal vs. external?

time neural activity

sensory data input stream self−generated internal dynamics

6

coupling to sensory input

self-generated internal dynamics

⊲ dHan (dense homogeneous associative network)

(0,1,2) (4,5,6) (6,7) (7,8,9,10) (2,3,4)

input dHan

input data stream

⊲ unrelated to internal dynamics

autonomous dynamics

sensory input

7

learning during sensible periods

competition internal dynamics ⇔ sensory input

1200 1400 time 0.2 0.4 0.6 0.8 1

xi

1200 1400

• 1

1

ri

1200 1400 0.5 1

∆ri

[A] [B] [C] [D] [E] [F]

activities growth rates sensory input

8

clique receptive fields

C C C C C C C C C C

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 −0.1 −0.2

1 4 2 9 5 7 8 3 6 10

9

cognitive computation

Introducing Cognitive Computation

Editor-in-Chief: Amir Hussain, PhD University of Stirling, Scotland, UK Honorary Editor: Igor Aleksander, PhD Imperial College, London, UK Advisory Board Chair: John Taylor, PhD

King’s College, London, UK

Volume 1, Issue 1, March 2009 ISSN: 1866-9956 (print) 1866-9964 (online)

New for 2009

Cognitive Computation is an international, peer-reviewed, interdisciplinary journal that publishes cutting-edge articles describing original basic and applied work involving biologically- inspired computational accounts of all aspects of natural and artificial cognitive systems. It provides a new platform for the dissemination of research, current practices and future trends in the emerging discipline of cognitive computation that bridges the gap between life sciences, social sciences, engineering, physical and mathematical sciences, and humanities. Its main purpose is to establish a forum for bringing different scientific communities together to discuss key issues and challenges in the emerging area of cognitive computation and to promote an interdisciplinary understanding of the diverse topics, including those related to perception, action, attention, learning and memory, decision making, language processing, communication, reasoning, problem solving, and consciousness aspects of cognition. Cognitive Computation considers original contributions using theoretical, computational, experimental and integrative studies in cognitive systems, including (but not limited to): artificial intelligence, neural networks, cognitive neuromorphic engineering and other hardware implementations, cognitive robotics, autonomous cognitive systems, neuroscience nanotechnology, self-organizing, swarm and immune systems, complex systems and control theory, and computational cognitive neuroscience, as well as submissions focusing on the development of latest research into practical applications.

for submission and further information, visit us online: springer.com/12559

• a new journal
• interdisciplinary
• beyond neurobiology,

applications

• http://www.springer.com/12559
• Springer, starting 2009

10

graduate level textbook

• The small world phenomenon in

social and scale-free networks

• Phase transitions and self-organized

criticality in adaptive systems

• Life at the edge of chaos and

coevolutionary avalanches resulting from the unfolding of all living

• Living dynamical systems

and emotional diffusive control within cognitive system theory

(Springer, 2008)

11

Frankfurt Cognitive System Platform

meta network of neural networks

• JAVA platform: class diagram
• flexibility: full on-the-fly architectural reconfiguration
• GUI (graphical user inferface): auto adaptive

12