S C A L E Artificial Life Weekend November 9-10, 1996

Participants

Name: Institution Status email address:

Chris Adami Cal Tech Faculty adami@krl.caltech.edu

John Batali UCSD Faculty batali@cogsci.ucsd.edu

Rik Belew UCSD Faculty rik@cs.ucsd.edu

Titus Brown Reed UG brown@krl.caltech.edu

Frances Edillo UCLA Grad S fedillo@ucla.edu

Johan Chu Cal Tech Grad jchu@cco.caltech.edu

Travis Collier Cal Tech UG travc@ugcs.caltech.edu

Nick Gessler UCLA Grad S gessler@anthro.sscnet.ucla.edu

Lucy Hadden UCSD Grad S hadden@cogsci.ucsd.edu

Mike Hamilton UCR Faculty michael.hamilton@ucr.edu

Tom Kammeyer UCSD Grad S tkammeye@cs.ucsd.edu

Mark Land UCSD Grad S mland@cs.ucsd.edu

Koji Morikawa UCLA Post Doc morikawa@biology.ucla.edu

Mike Oliphant UCSD Grad S oliphant@cogsci.ucsd.edu

Charles Ofria Cal Tech Grad S charles@krl.caltech.edu

Filippo Menczer UCSD Grad fil@cs.ucsd.edu

Joao Munoz UCLA Grad S munoz@biology.ucla.edu

Chris Rosin UCSD Grad S crosin@cs.ucsd.edu

Brian Skyrms UCI Faculty bskyrms@uci.edu

Chuck Taylor UCLA Faculty taylor@biology.ucla.edu

Peter Thumfort UCLA Post Doc thumfort@biology.ucla.edu

Saturday, November 9, 1996

Misc Notes:

send Batali Arita/Taylor

Farmer Rosetta to Brian Skyrmes

Vavilov to Rick

course at UCLA - mLeanie and rik and Darry D. Davis

Batali

paper for takaya send email to takaya

Yanko and stein - for Takaya (on line)

in animals to animats 2

corect citations on the web

Pass-Through - Introduction

Hamilton

33 reserves now

5,000 per year here

Databases for diversity

GIS to reality after recent fire of 10,000 acres - $1M for Mike

his model worked well

John Batali

last time- described motivation systems (pleasure/pain) guide learning

it worked but biases were different than real life, actually better - Evol Comput in press

also evolution of language

Lucy Hadden -

with John, animat-style work. How to build domestic chicks, filial imprinting

Frances Edillo-

Charles Offria-

Avida system, coding, tools, cross-platform

Brian Skyrmes-

evolutionary dynamics of interactions, game theory

investigating evolution of strategy in bargaining games

finished book - evolution of the social contract - from Cambridge

The Dynamics of Norms - good book

Joao Munoz-

hierarchical natural selection - how to address it?

Mike Oliphant-

language, what is needed in a learning mechanism for a population to establish a lexicon and maintain it over time

Thomas Kameyer-

Machine learning, GA for grammatical inference

how take a GA for morphogenesis

amount of testing per individual, how that affects dynamic of ga

Chris Rosen-

GA on coevolution

competitive coevolution of test-cases

Mark Land-

local search in GA does optimization

diff Lamarkian and non-Lamarkian evolution

Filippo Menczer-

LEE - how different forms of environment affect evolution

coming out in Adaptive Behavior

working now on more applied approach -evolve agents for web searching

Travis Collier-

working with Chris Adami

using Avida to simulate bacterial populations - directed evolution

Titus Brown-

UG at Reed, working with Mark Bedau. Cross-model comparison

Avida, tierra, Packard's bug

Pere Bak - how nature works

hire UGs for about 10 hrs per week is the best way

wrote Avida in Java, does truly work cross-platform, but is not easy

Nick Gessler-

Workshop in Computational Social Science, Feb 28, March 1-2.

willing to put up a web site for this

Peter Thumfort-

oxygen diffusion and self-regulation

"genome project is like dancing around the golden calf"

Chris Adami-

Change in attitude at caltech

Lectures in intro biology at Cal tech

discussion grous - e.g. computational physics and biology

e.g pattern formation -- Fleischer-Barr

$2.5 million for predictive models in computational biology - development

Modeling Nature with Cellular automata with Mathematica -

outstanding with D Rom - Gaylor and Nishidate - Springer-Verlag

Johann Chu-

Rik -

Lander article Genome issue of Science - stresses need to understand function

Genome project colors how we approach biology

Would an ecology project help with developing a community here?

Potential topics:

1. Communication (from last year)

2. Rosetta Stone

(subset that gives meaningful result)

3. Adaptive Surfaces

4. Thermodynamic definition of life

5. Complex Adaptive Systems for media/ted tools

6. Mike's data - ecological database for alife work

7. Blife project

8. Joint research

9. web presence

10. cultural evolution

11. Genome project colors how we approach biology

Chose among those for Schedule

AM

Communication

Adaptive surfaces

PM

Cultural evolution

thermodynamic definition of life +self-modifying code

adaptive surfaces

Night

Mike -- demo

beyond genome

Tomorrow

Joint research

web presence

Discussion of Adaptive Surface Group

Chuck Taylor, Chris Adami, Mike Hamilton, Johan Chu, Travis Collier, Filippo Menczer

1.Types of Landscape

A. Kauffman

a function on bit-springs

a string of length L, 2^L such strings

to assign a fitness on each,

contribution of each bit adds to a total

make a table of such strings, assign a random number to each

add up the values, total is fitness

B. Lewontin (Started with Wright)

each axis represents an allele frequency

and mean population fitness is a real number over RN

B. Eigen

defined on [nucleotide string] abstract replicator

strings of code that self -replicate

each [nucleotide string] has a fitness

is the number of offspring per unit time

meant for RNA world

C. Wright

1931--

D. Implicit Landscapes-Adami

space of all possible strings of all lengths

Each genotype has its own replication rate,

fitness defined by replication rate of each string as it is embedded in the population of other strings

if change allele frequencies, then the fitness will change

from this you can create Lewontin fitness map

but there is no function where you plug in string and get out a real number

you must measure it in a system

so you have no fitness independent of what is around it

basically it must be empirically measured

does not work for diploid organism or organisms with recombination

E. Fitness Function - Parisi

each "genome" is a vector of real numbers (haploid so far)

There is a black box tht represents a population of such vectors

put in our "genome" and read out a real number (they call it fitness, but it is not really fitness as geneticists use it, rather it is a competitive value)

take all such competitive values, rank them along the x axis, then

the fitness comes from a step function over this domain, like with threshold selection on GAs. The threshold is always relative to the pop -- e.g. the top 10% of the pop.

{we did not get to the following issues}

2. advantages of Each as a metaphor

individual

3. Theoretical problems with each as a metaphor

4. Empirical Concerns

In the afternoon we attempted to fill in a chart:

Model diploid/haploid freq dep? poymorphism? etc

kauffman

Lewontin

modified

Eigen

Schuster

Matiss

general

Implicit

parisi (GA)

There were a lot of differences among these, and it was clear that some of the restrictions were not fully appreciated. We felt this should be followed up.