From Amoeba to Cognition Frankfurt Institute of Advanced Studies April 16, 2003 Christoph von der...

From Amoeba to Cognition Frankfurt Institute of Advanced Studies

April 16, 2003

Christoph von der Malsburg Institut für Neuroinformatik

und Fakultät für Physik und AstronomieRuhr-University Bochum, Germany

andComputer Science Departmentand Program in Neuroscience

University of Southern CaliforniaLos Angeles

Amoeba

Euglena

Repertoire of single-celled animals 1

Metabolism

• Production, transformation and breakdown of molecules

• Synthesis of molecules under genetic control

• Regulation, e.g., of ionic concentrations

• Transport of molecules, inside, in and out of cell

• Electrical “behavior”

• Circadian rhythm

Reproduction

Repertoire of single-celled animals 2

Behavior

•Sensing (light, sound, chemical milieu)

•Self-shaping (pseudopodia, mitosis)

•Motility, esp. chemotaxis

•Feeding: ingestion and digestion

•Aggression, flight

•Signalling

•Collaboration (e.g., slime mold, biofilms)

Amoeba aggregation 2

Spiral waves

Ants

Neuron 1

Neuron 2

Synapse

The Ontogenetic “Riddle”

• Information content of the genome: 109 bits

• Information content of the brain’s wiring: 1016 bits

(1010 neurons, hence ld 1010 = 33 bits per connection,

times 1015 synapses = 1016 bits of information)

• Solution: genetically controlled self-organization

Rettec anatomical schemaA Model for the Ontogenesis of Retinotopy

(Willshaw and Malsburg, 1976)

Rettec functional schma

• Chemotaxis

• Synaptic plasticity controlled by electrical signals

Hebbian Plasticity

Correlation-controlled Synaptic Plasticity

(“Hebbian Plasticity”)

Time10 sec

Meister

(Prenatal ferret retina, M. Meister et al.)

Network Self-Organization

Network Signals

Signal Dynamic

Synaptic Plasticity

Rettec functional schma

Rettec principle 2

Rettec development

Visual system schema

Levay stripes

Binoc 1 A Model for the Ontogenesis of Ocularity Domains (Biol. Cybernetics, 1977)

Binoc 2

H&W orient

Devalois 2

73 projectionA model for the development of orientation-

specific neurons (Kybernetik, 1973)

Retina

Cortex

Connection Strength

73 stimuli Retinal Stimuli

Meister

(Prenatal ferret retina, M. Meister et al.)

73 cell 70 Re-organization of a cortical receptive field

73 cortex post

73 orientmap

Devalois 1

Gabors

Olshausen-and Field: Schema

Development of connections strengths Φi(x,y) under 2 constraints:

• Preservation of information (ability to reconstruct)

• Sparsity

Natural images

Olshausen-Field Gabors

Points of Conclusion:

• Retinotopy, orientation specificity as paradigmsof network self-organization and CNS ontogenesis

• Ontogenesis of CNS and cellular repertoire

• Amount of genetic information

Invariant object Recognition(As paradigm of a cognitive function)

image model

van Essen

Rubfig 1Image Domain Model Domain

Model Window

Object recognition

Rubfig 2Image Domain Model Domain

Model Window

Objection recognition 2

Temporal binding

Temporal binding

Rapid, Reversible Synaptic Plasticity

Time10 msec

Network Self-Organization

Network Signals

Signal Dynamic

Synaptic Plasticity

Image-to-jets

Maryl-representation

2D mapping formation

Face recognition rates

Model Probe Size Recognition rate *

Other systems

frontal Diff expressionlarge transform

124 85%

frontal Diff expressionsmall transform

124 96.8% 98% (=245/250)(Wiskott et al 97)

frontal 30° rotation in depth

110 93.6% 66.4% (=73/110)(Wiskott & Malsburg 96)

* After 3 iterations

Points of Conclusion:

• Evolution as a game of varying the eurkaryote’s repertoire

• Ontogenesis as a refinement of old cellular behavioral patterns

• reproduction, differentiation

• cellular migration, chemotaxis

• chemical signalling, reaction-diffusion patterns

• putting out of “pseudopodia”

• Brain function as a fast version of the same game again

• Network Self-Organization the central process

Outlook

• The flexibility of the human brain shows that fundamental principles are at work

• Similar conclusions may be drawn from the rapid development of human society

• Elucidating the general principles of organization is the challenge of our times

• This issue has at present no academic home

Molecular Biology

The Software Crisis

NIST Study 02: yearly US loss due to SW failure: $60 Billion

Human:

Detailed Communication

Machine:

Creative Infrastructure: Goals, Methods, Interpretation, World Knowledge, Diagnostics

Algorithms: deterministic, fast, clue-less

Algorithmic Division of LaborAlgorithmic

DOL

Human:

Loose Communication

Machine:

Goal Definition

Creative Infrastructure:Goals, Methods, Interpretation, World Knowldege, Debugging

Data, „Algorithms“

Organic ComputersOrganic

Computers

Self-Organization in Need of Development

The ideas of self-organization have created a revolution, but they are now in need of forceful further development!

Underdeveloped aspects:

• Control of the control parameters (Ashby’s super-stability)

• Explicit representation of goals

• Cascades of organization (description of unfolding systems)

• Escaping geometry (e.g., network self-organization)

Physics to the Rescue!!• Physics has a proven track-record of understanding complex

phenomena on the basis of simple paradigms and principles

• Physics is in possession of highly relevant methodology(statistical mechanics, systems of non-linear differential equations)

• Physics has a very successful system of education

• Physics is on the look-out for a new application field