Download presentation
Presentation is loading. Please wait.
1
Immunocomputing and Artificial Immune Systems
By Daniel Reeves
2
Copyright 2007 Daniel Reeves
Overview Introduction Immune System Immune system characteristics Artificial Immune Systems (AIS) AIS algorithms and applications Questions Copyright 2007 Daniel Reeves
3
Copyright 2007 Daniel Reeves
Introduction Immune System Immunocomputing Generation of new software and algorithms based in principles of immune system components (proteins, immune networks) leading to concept of immunocomputer. Immunocomputer Artificial Immune Systems Copyright 2007 Daniel Reeves
4
Noteworthy AI related Immune System Characteristics
Pattern recognition Replication Genetic Alteration Multi-level response capabilities Swarm Behavior Networks Copyright 2007 Daniel Reeves
5
Immune System Physiology
Primary (creation) Bone Marrow Thymus Secondary (interaction) Spleen Appendix Tonsils Lymphatic Nodes and Vessels Peyer’s Patches Copyright 2007 Daniel Reeves
6
Immune System Physiology
Immune System Cells B-Cells T-Cells Foreign Cells Pathogens Antigens Epitopes Copyright 2007 Daniel Reeves
7
Immune System – Response Systems
Innate Immune System 1st line of defense General recognition Holds Pathogens at bay Adaptive Immune System Recognition Memory Response Copyright 2007 Daniel Reeves
8
Primary Immunological Principles
Pattern Recognition Adaptive immune responses Self/non-self discrimination Immune Network theory Danger Theory Copyright 2007 Daniel Reeves
9
Recognition and Binding
Antigens and Epitopes Antibodies and Immunoglobulin Affinity Copyright 2007 Daniel Reeves
10
Adaptive Immune Responses
Clonal Selection Theory Affinity maturation Mutation Copyright 2007 Daniel Reeves
11
Self/Non-self Discrimination
Self and non-self Negative selection Co-stimulatory signals Copyright 2007 Daniel Reeves
12
Copyright 2007 Daniel Reeves
Immune Network Theory Learning Structure Dynamics Metadynamics Self-tolerance Copyright 2007 Daniel Reeves
13
Copyright 2007 Daniel Reeves
Danger Theory Many types of cells (self and non-self) Dangers to immune system from both Difficult to tell what is and is not dangerous Suggests: “Immune system is more concerned with damage (preventing destruction) than foreignness” And: “antigen presenting cells are activated by danger signals given off by injured or damaged cells.” Copyright 2007 Daniel Reeves
14
Artificial Immune Systems
Definition “adaptive systems inspired by theoretical and experimental immunology with the goal of solving problems” Immune engineering Representation Evaluation Mechanisms Adaptation Procedures Copyright 2007 Daniel Reeves
15
Immune Engineering - Representation
Immune Response by Shape Model Shape Space Real-valued Integer-valued Hamming Symbolic Copyright 2007 Daniel Reeves
16
Immune Engineering – Evaluation Mechanisms
Affinity Measures Real-valued vectors Euclidian & Manhattan Distances Integer Strings Hamming Strings Hamming Distances Binary-Hamming shape space Affinity Threshold Cross-reactivity threshold Copyright 2007 Daniel Reeves
17
Copyright 2007 Daniel Reeves
AIS Algorithms Bone Marrow Negative Selection Clonal Selection Continuous immune network Discrete immune network Copyright 2007 Daniel Reeves
18
Copyright 2007 Daniel Reeves
Bone Marrow Models Generate populations of cells Genetic Libraries Genotype and Phenotype Knowledge Storage Applications Evolution of Genetic encoding of antibodies Antibodies for Job scheduling Copyright 2007 Daniel Reeves
19
Bone Marrow Pseudocode
Pseudo-Code (antibody generation) procedure[Ab] = gene_library(N) initialize L for i from 1 to N for j from 1 to n geneIndex = rand(L[1]) Ab[i][j] = L(j,geneIndex) end for end procedure Copyright 2007 Daniel Reeves
20
Negative Selection Algorithms
Define set of anomaly detectors Applications Network Intrusion Detection Breast Cancer Diagnosis Copyright 2007 Daniel Reeves
21
Negative Selection Algorithms
Pseudo-code (non-self generation) Generate random detectors Generate random set of possible detectors Evaluate Affinity Compare Random detectors to set of self patterns Accept or Reject If there is a match, reject the detector If there is not a match, add detector to set of non-self detectors Copyright 2007 Daniel Reeves
22
Clonal Selection and Affinity Maturation
Features System responds to subset of cells Affinity level to offspring generation relation Affinity level to offspring mutation relation Algorithms Forrest’s Algorithm ClonalG Applications Network Security Job-Shop scheduling Copyright 2007 Daniel Reeves
23
Copyright 2007 Daniel Reeves
Forrest’s Algorithm Pseudo-code Initialize population Match antibodies & antigens for affinity Score affinity of antibodies Evolve antibody repertoire (GA) Fitness measure Choose antigen and sample of antibodies Match each antibody to antigen Add match score to antibody Repeat for many antigens Copyright 2007 Daniel Reeves
24
Copyright 2007 Daniel Reeves
ClonalG Similar to Forrest’s algorithm but Does have: affinity proportional selection and mutation Doesn’t have: locality of immune response Pseudocode Initialize Antigenic Presentation Affinity Evaluation Clonal Selection and expansion Affinity Maturation Metadynamics Cycle Copyright 2007 Daniel Reeves
25
Artificial Immune Networks
Two models Continuous Discrete Immune system dynamic Applications Recommender System Data Compression and Clustering Copyright 2007 Daniel Reeves
26
Continuous vs. Discrete
Ordinary Differential Equations Concentrations of antibodies Used for simulation of immune networks Iterative Process or difference equation Variations in antibody number and structure Used for solving problems Copyright 2007 Daniel Reeves
27
Natural World AIS Mapping
Cells and Molecules Affinity Fitness Bone-marrow models Affinity Function Hyper mutation Affinity maturation Clonal selection Negative Selection Immune network Attribute Strings Degree of match Data structure quality Generate structures Quantify affinities Maintain population Promotes learning Interaction with antigens Self/Non-self detectors Performs dynamics Copyright 2007 Daniel Reeves
28
Copyright 2007 Daniel Reeves
AIS Example - TSP Problem: Traveling Salesman Problem Distribution of points in 2D plane Visit each point once and make one circuit Euclidian distance for path length Bone Marrow Algorithm (BM) for initial candidate solutions (genes) ClonalG Algorithm (CG) for generating better solutions (Affinity Maturation) Copyright 2007 Daniel Reeves
29
Copyright 2007 Daniel Reeves
TSP Example Copyright 2007 Daniel Reeves
30
Copyright 2007 Daniel Reeves
Representation Genes: Strings of four x/y pairs Gene Libraries: Collections of genes Antibodies: Groups of five genes Affinity Equation: Repair Function: Remove duplicate points and add in missing Copyright 2007 Daniel Reeves
31
Copyright 2007 Daniel Reeves
Bone Marrow Initialize gene libraries Randomly chose point & 3 closest neighbors Build 5 different libraries of genes Build antibody population from gene libraries Grab one gene from each library Repair bad antibodies Use repair function to repair antibodies Copyright 2007 Daniel Reeves
32
Copyright 2007 Daniel Reeves
ClonalG Initial Population: Candidate solutions from bone marrow algorithm Affinity Evaluation, Clonal Selection and Expansion: Choose best candidate solutions from BM Affinity Maturation: Create clones to mutate based on affinity level Mutate created clones based on affinity level Metadynamics: Kill off poor cells and get new cells from BM Repeat Process until stopping criterion Copyright 2007 Daniel Reeves
33
Questions?
34
Copyright 2007 Daniel Reeves
TO DO Find Pictures of Immune system components Copy table translation from natural to artificial Copyright 2007 Daniel Reeves
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.