1 / 27

Classifier Systems

Classifier Systems. Overview. Introduction and problem overview Architecture Component details Track a specific example Summary. The Learning Classifier System. Rule-based knowledge discovery and concept learning tool

jkerns
Download Presentation

Classifier Systems

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Classifier Systems

  2. Overview • Introduction and problem overview • Architecture • Component details • Track a specific example • Summary

  3. The Learning Classifier System • Rule-based knowledge discovery and concept learning tool • Operates by means of evaluation, credit assignment, and discovery applied to a population of “chromosomes” (rules) each with a corresponding “phenotype” (outcome)

  4. Components of a Learning Classifier System • Performance • Provides interaction between environment and rule base • Performs matching function • Reinforcement • Rewards accurate classifiers • Punishes inaccurate classifiers • Discovery • Uses the genetic algorithm to search for plausible rules

  5. Knowledge Representation • Classifiers • IF-THEN rules • Condition=“genotype” • Action=“phenotype” • Strength metric • Encoded as bit strings or numerics • Population • Fixed size collection of classifiers

  6. Low-level knowledge representation:The Classifier • Taxon is analogous to a condition (LHS) of an IF-THEN rule • Action bit is analogous to an action (RHS) of an IF-THEN rule • Strength is an internal fitness function

  7. Problem Multiplexer Example Perfect Rule Set

  8. Classifier System (C.S) • Learn simple string rules in an arbitrary environment • A classifier is a simple string rule • Components • Rule and Message System • Apportionment of credit system • Genetic Algorithm

  9. Overview • General organization of a classifier system • performance system: rule based, message-passing, highly standardized, and highly parallel • credit assignment: bucket brigade algorithms • rule discovery: genetic algorithms

  10. Definition of the basic elements • input interface: translate the current state of the environment into standard messages • classifiers (the rules used by the system): define the system’s procedures for processing messages • message list: contain all current messages (those generated by the input interface and those generated by satisfied rules) • output interface: translate some messages into effector actions that modify the state of the environment

  11. Classifier Systems (2) • Basic parts of a classifier system

  12. Execution cycle • Step 1. Add all messages from the input interface to the message list. • Step 2. Compare all messages on the message list to all conditions of all classifiers and record all matches (satisfied conditions) • Step 3. For each set of matches satisfying the condition part of some classifiers, post the message satisfied by its action part to a list of new messages. • Step 4. Replace all messages on the message list by the list of new messages. • Step 5. Translate messages on the message list to requirements on the output interface, thereby producing the system’s current output. • Step 6. Return to Step 1.

  13. Rule and Message System • Production system • Fixed size representation for rules • Parallel activation • Rating of a rule by an information-based economy • <message>::= { 0, 1} l • <classifier>::= <condition>:<message> • <condition>::={0, 1, #}l

  14. Which classifier to choose? • Bucket Brigade Algorithm • For ranking or rating individual classifiers • Classifiers buy and sell the right to trade information (information-based economy) • Auction house and clearing house • If a classifier matches a message, it participates in an auction. • The bid (B) is proportional to its strength (S) • Once activated the winner pays its bid to other classifiers which also matched the message

  15. Which classifier to choose?(contd…) • Notation • S = Strength • P = Payment • T = Tax • R = Reward • Cbid = Bid Coefficient • The ith classifier strength (at time step t) Si(t+1) = Si(t) – Pi(t) – Ti(t) + Ri(t) • Bid Bi = Cbid * Si • Tax Taxi = Ctax * Si • Effective Bid EBidi = Bi + N (σbid) • In terms of strength S(t+1) = S(t) – Cbid*S(t) – Ctax*S(t) + R(t)

  16. Generating better rules • Bucket brigade algorithm evaluates rules and decides among competing alternatives. • We could still inject new (possibly better rules), so use a Genetic Algorithm (GA) • A classifier’s strength (S) is used as its fitness • Similar to the simple genetic algorithm • Entire population is not replaced at the next generation (Generation gap ) • GA period (epoch) • Number of time steps between GA calls • Time step = rule-message cycle • Crowding to maintain diversity • Mutation over a ternary alphabet {1, 0, # }

  17. Generating better rules • Selection is performed using roulette-wheel selection • The GA is run according to the GA Period or when conditioned on particular events (lack of match or poor performance)

  18. T= 0 C.S in action (1) Strength (S) Messages (Msg) Match (M) Bid (B) CBid = 0.1 CTax = 0.0

  19. C.S in action (2) T= 1 Strength (S) Messages (Msg) Match (M) Bid (B) CBid = 0.1 CTax = 0.0

  20. C.S in action (3) T= 2 Strength (S) Messages (Msg) Match (M) Bid (B) CBid = 0.1 CTax = 0.0

  21. C.S in action (4) T= 3 Strength (S) Messages (Msg) Match (M) Bid (B) CBid = 0.1 CTax = 0.0

  22. C.S in action (5) T= 4 Strength (S) Messages (Msg) Match (M) Bid (B) CBid = 0.1 CTax = 0.0

  23. C.S in action (6) T= 5 Strength (S) CBid = 0.1 CTax = 0.0

  24. Are these rule-sets the same?

  25. Multiplexer Example • Default Hierarchy • General rules cover general conditions and specific rules cover exceptions • Parsimony • Fewer rules • Enlargement of the solution set • While the problem space remains the same

  26. Summary • A classifier is a simple string rule • Classifier System • rule-message system, • apportionment of credit mechanism • GA • Advantages of CS • rules are simple • use fixed length representation • parallel activation • operate in an information-based economy

  27. Thank You Questions ?

More Related