Case Injected Genetic Algorithms Sushil J. Louis Genetic Algorithm Systems Lab (gaslab) University of Nevada, Reno

Collaborators  Vinod Gandikota  Igor Golovkin  Xiaohua Liu  Andrew Murray  Rikun Tang  Indira Vinjamuri  Yongmian Zhang

Outline  Motivation  What is the technique? Genetic Algorithm and Case-Based Reasoning  Is it useful? Combinational Logic Design Strike Force Asset Allocation TSP Scheduling  Conclusions

Genetic Algorithm  Non-Deterministic, Parallel, Search  Poorly understood problems  Evaluate, Select, Recombine  Population based search Population member encodes candidate solution Building blocks combine to make progress More resistant to local optima Iterative, requiring many evaluations

Motivation  Deployed systems are expected to confront and solve many problems over their lifetime  How can we increase genetic algorithm performance with experience?  Provide GA with a memory

Case-Based Reasoning  When confronted by a new problem, adapt similar (already solved) problem’s solution to solve new problem  CBR  Associative Memory + Adaptation  CBR: Indexing (on problem similarity) and adaptation are domain dependent

Case Injected Genetic AlgoRithm  Combine genetic search with case-based reasoning  Case-base provides memory  Genetic algorithm provides adaptation  Genetic algorithm generates cases Any member of the GA’s population is a case

System

Related work  Seeding:Koza, Greffensttette, Ramsey, Louis  Lifelong learning: Thrun  Key Differences Store and reuse intermediate solutions Solve sequences of similar problems

Combinational Logic Design  An example of configuration design  Given a function and a target technology to work with design an artifact that performs this function subject to constraints Target technology: Logic gates Function: Parity checking Constraints: 2-D gate array

Encoding

Encoding

Parity Input3-bit Parity3-1 problem

Problem similarity

Lessons  Storing and Injecting solutions may not improve solution quality  Storing and Injecting partial solutions does lead to improved quality

OSSP Performance

Which cases to inject?  Problem distance metric (Louis ‘97) Domain dependent  Solution distance metric Genetic algorithm encodings Binary – hamming distance Real – euclidean distance Permutation – longest common substring …

Solution Similarity

Periodic Injection Strategies  Closest to best  Furthest from worst  Probabilistic closest to best  Probabilistic furthest from worst  Randomly choose a case from case-base  Create random individual

Setup  50, 6-bit combinational logic design problems  Randomly select and flip bits in parity output to define logic function  Compare performance Quality of final design solution (correct output) Time to this final solution (in generations)

Parameters  Population size: 30  No of generations: 30  CHC (elitist) selection  Scaling factor: 1.05  Prob. Crossover: 0.95  Prob. Mutation: 0.05  Store best individual every generation  Inject every 5 generations (2^5 = 32)  Inject 3 cases (10%)  Multiple injection strategies Averages over 10 runs

Problem distribution

Performance - Quality

Performance - Time

Injection Strategies

Solution distribution

Strike force asset allocation  Allocate platforms to targets  Dynamic Changing target priority Battlefield conditions Popup Weather …

Factors in allocation  Pilot proficiency  Asset suitability  Priority  Risk Route Other assets (SEAD) Weather

Maximize mission success  Binary encoding  Platform to multiple targets  Target can have multiple platforms  Dynamic battle-space Strong time constraints

Setup  50 problems.  10 platforms, 40 assets, 10 targets  Each platform could be allocated to two targets  Problems varied in risk matrix  Popsize=80, Generations=80, Pc=1.0, Pm=0.05, probabilistic closest to best, injection period=9, injection % = 10% of popsize

Results

TSP  Find the shortest route that visits every city exactly once (except for start city)  Permutation encoding. Ex:  Similarity metric: Longest common subsequence (Cormen et al, Introduction to Algorithms)  50 problems, move city locations

TSP performance

Scheduling  Job shop scheduling problems  Permutation encoding (Fang)  Similarity metric: Longest common subsequence (Cormen et al, Introduction to Algorithms)  50 problems, change task lengths

JSSP Performance (10x10)

JSSP Performance (15x15)

Summary  Case Injected Genetic AlgoRithm: A hybrid system that combines genetic algorithms with a case-based memory  Defined problem-similarity and solution- similarity metrics  Defined performance metrics and showed empirically that CIGAR learns to increase performance for sequences of similar problems

Conclusions  Case Injected Genetic AlgoRithm is a viable system for increasing performance with experience.  Improving one or both of Quality of solution found – highest fitness individual Number of generations needed to find this solution  Repeated injection based on similarity  Syntactic similarity measures suffice Hamming distance Longest Common Sub-string for permutation encoding

Implications  Implications for system design Increases performance with experience Generates cases during problem solving Long term navigable store of expertise Problem analysis by analyzing case-base