Background “Structurally dynamic” cellular automata (Ilachinski, Halpern 1987) have been shown to simulate biological functions with emergent behavior.

Slides:



Advertisements
Similar presentations
Artificial Neural Networks
Advertisements

Standard II- Cell Processes
1 Machine Learning: Lecture 4 Artificial Neural Networks (Based on Chapter 4 of Mitchell T.., Machine Learning, 1997)
Artificial Neural Networks (1)
Diffusion and Osmosis.
Artificial Neural Networks ML Paul Scheible.
Prénom Nom Document Analysis: Artificial Neural Networks Prof. Rolf Ingold, University of Fribourg Master course, spring semester 2008.
1 Chapter 11 Neural Networks. 2 Chapter 11 Contents (1) l Biological Neurons l Artificial Neurons l Perceptrons l Multilayer Neural Networks l Backpropagation.
Prénom Nom Document Analysis: Artificial Neural Networks Prof. Rolf Ingold, University of Fribourg Master course, spring semester 2008.
Neural Networks Lab 5. What Is Neural Networks? Neural networks are composed of simple elements( Neurons) operating in parallel. Neural networks are composed.
Cellular Transport Unit Passive Transport = movement of substances across the cell membrane without any input of energy by the cell.
Bellringer List the two things that make up the cell membrane.
Cell Function – Cells and their Environment
Presentation on Neural Networks.. Basics Of Neural Networks Neural networks refers to a connectionist model that simulates the biophysical information.
Neural Networks Ellen Walker Hiram College. Connectionist Architectures Characterized by (Rich & Knight) –Large number of very simple neuron-like processing.
Introduction to Artificial Neural Network Models Angshuman Saha Image Source: ww.physiol.ucl.ac.uk/fedwards/ ca1%20neuron.jpg.
Movement of particles across the cell membrane without using energy
1 Chapter 11 Neural Networks. 2 Chapter 11 Contents (1) l Biological Neurons l Artificial Neurons l Perceptrons l Multilayer Neural Networks l Backpropagation.
Passive Transport Diffusion Osmosis Biology Unit - Regulation.
Pencil-and-Paper Neural Networks Prof. Kevin Crisp St. Olaf College.
Cellular Transport 7.4. Diffusion Baking cookies aroma Baking cookies aroma Particles in G, L, S are in constant motion Particles in G, L, S are in constant.
Cells and Their Environment. Hooke and van Leeuwenhoek postulated: 1. All living things are composed of cells 2. Cells are the basic unit and function.
Homeostasis & Transport. Section 1: Diffusion & Osmosis.
September Bound Computation for Adaptive Systems V&V Giampiero Campa September 2008 West Virginia University.
CAP6938 Neuroevolution and Developmental Encoding Basic Concepts Dr. Kenneth Stanley August 23, 2006.
Cell Transport Osmosis and Diffusion.  Particles in constant motion  Run into each other and randomly spread out  Particles move from an area of high.
M. Tweedie College/Honors Biology. First, Some Vocabulary!!!! Concentration gradient: the difference between the HIGH and LOW concentration areas (tells.
Cellular Transportation. Diffusion TO High Concentration Low Concentration.
Adaptive Hopfield Network Gürsel Serpen Dr. Gürsel Serpen Associate Professor Electrical Engineering and Computer Science Department University of Toledo.
Cellular Transport.
Neural Networks Presented by M. Abbasi Course lecturer: Dr.Tohidkhah.
Model-based learning: Theory and an application to sequence learning P.O. Box 49, 1525, Budapest, Hungary Zoltán Somogyvári.
Image Source: ww.physiol.ucl.ac.uk/fedwards/ ca1%20neuron.jpg
Diffusion and Osmosis Transport- Passive or Active Passive transport-the movement of materials across a cell membrane without the expenditure of cell.
Cellular Processes *Transportation of Substances In/Out of Cell (Active vs. Passive Transport)  Unit Three *Respiration  Unit Three *Photosynthesis 
Materials move across membranes because of concentration differences.
Cell Transport.  The molecules can move freely in the membrane.  They are constantly renewed during a cell's life span.  Grow with the growth of.
Intelligent Control Methods Lecture 14: Neuronal Nets (Part 2) Slovak University of Technology Faculty of Material Science and Technology in Trnava.
7.4 – Cellular Transport. Remember from 7.2….. Selective Permeability – substances do NOT enter the cell indiscriminately. –Nonpolar molecules tend to.
Cells and Their Environment. Cells interact with their environments!
Cellular Functions All the different parts of the cell are important to the life of the cell. Cellular activities require energy. All cells are able to.
A Presentation on Adaptive Neuro-Fuzzy Inference System using Particle Swarm Optimization and it’s Application By Sumanta Kundu (En.R.No.
7 th Grade Science Week of October 18 Science Starters.
1 Neural Networks MUMT 611 Philippe Zaborowski April 2005.
Section 4: Cellular Transport
Cell Transport Chapter 7.3.
Cellular Transport Notes
3.4 Diffusion and Osmosis KEY CONCEPT Materials move across membranes because of concentration differences.
Cellular Transport Notes
Section 4: Cellular Transport
Passive Transport: Simple Diffusion & Osmosis
Cellular Transport Notes
By: Diana Bivens Modified by: Kerri Shrestha
Take out notebook and open to next available page. Write:
Section 4: Cellular Transport
The Cell Membrane A. Fluid-Mosaic Model – a semi- permeable membrane composed of a phospholipid bilayer with embedded proteins.
Osmosis.
Cellular Transport Notes
Artificial Intelligence Chapter 2 Stimulus-Response Agents
Cellular Transport Notes
Types of Cellular Transport
Into to Cellular Transport
Movement of particles across the cell membrane without using energy
The Cell Membrane and Transport
Chapter 7 Cellular Structure and Function
Moving Cellular Materials
BELLWORK What are the three organelles that only plants have?
Cellular Transport.
Types of Cellular Transport
Moving Cellular Materials
Presentation transcript:

Dynamic cellular automata control using multiple recurrent neural nets and unsupervised learning

Background “Structurally dynamic” cellular automata (Ilachinski, Halpern 1987) have been shown to simulate biological functions with emergent behavior “Agent-based” models show how complex macro behavior in a population emerges from simple individual rules My model is not structurally dynamic, just dynamic I used neural nets because they allow for a broad search space of cellular behavior, given some fixed environmental challenges.

Cellular automata Written in Typescript, a JavaScript superset with types Uses JavaScript library of neuron nets, Synaptic. Allows for arbitrary “projecting of layers” Composable

Demo Jamiesonwarner.com/plants

Setting The cellular automata is an M by N grid of cells. Each cell is a “fluid vector” {f1, …, fn}, each fi is a scalar. The “plant” is the set of living cells in the automata. Some dynamic systems simulated: diffusion, osmosis, and biological functions: photosynthesis, active transport, cell metabolism, cell death, and cell reproduction

Control model A set of m actions is defined. Each tick, every cell executes one of the m actions. Each living cell is assigned a cell type Each cell type x action pair is assigned a neural net: n inputs, where n is the number of fluids 1 output, representing the decision weight for this action Every tick, the decision weight for every action is computed on every cell, and the action chosen is result of the softmax function over the decision weights

Action directionality I wanted action choices to be symmetric left and right To accomplish this, whenever a “directional” action is performed, First the fluid gradient is computed at the performing cell (2 x n-dimensional) Then the direction computed is a constant matrix times the fluid gradient

Problem encountered: balancing Some early environmental challenges made it “too hard” for plant growth. For instance early plants would die of dehydration immediately upon reaching the air because the air had lower humidity levels => death by passive transport As a solution I decreased passive transport into the plant, and gave plants the ability to regulate water pressure by pumping fluid from cell to cell.

Training Random search over all weights of the neural nets (BIG search space) Fitness is defined as the number of living cells after G=100 ticks

Extensions Relax adjacency constraints to make a “structurally dynamic” cellular automata (Ilachinski, Halpern 1987) Get rid of cell types in favor of finer control over the fluid vector Every dimension of the fluid vector (past dimension K) gets its own neural net The value fi , i >K, is the sum of its current value and the output of the neural net Use simulated annealing search

Project url: https://github.com/jamiesonwarner/springfever