1 / 14

Updates - PowerPoint PPT Presentation

  • Uploaded on

Updates. September 24 th , 2013 Erik Fredericks. Overview. Updates from previous meeting Literature review on local optima. Updates. Removed incremental evaluation of pre/post conditions Left in check for valid/invalid transforms

I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
Download Presentation

PowerPoint Slideshow about ' Updates' - dalit

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.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.

- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript


September 24th, 2013

Erik Fredericks


  • Updates from previous meeting

  • Literature review on local optima


  • Removed incremental evaluation of pre/post conditions

    • Left in check for valid/invalid transforms

  • Added in secondary fitness function to increase depth of tree

    • Also increased maximum tree depth

  • Reintroduced crossover

    • Mutation rate: 25%

    • Crossover rate: 50%


  • Moving away from proper solutions

    • Invalid transformation chains


      • VOID2INDEX(float-array)

  • Appears to be a flaw in code, not in approach

    • Still hunting this bug down


  • Diversity is back from crossover operations

    • Generational run takes much longer now than before

  • Cannot yet comment on performance of algorithm until glitch is fixed

Local optima paper reviews
Local Optima Paper Reviews

  • Novelty search in GE [Urbano2013]

    • Get out of local optima in Santa Fe Ant Trail problem

      • Deceptive problem

    • No archive added as experiments showed it did not help

    • Results show that novelty search outperforms standard GE

Local optima paper reviews1
Local Optima Paper Reviews

  • Other GE approach

    • Grammatical herding [Headleand2013]

      • Swarm-based heuristic

      • Treats environment as solution space and ‘herds’ solutions towards high-fitness areas

      • Contains:

        • Herd – standard population of individuals

        • Betas – subset of fittest agents to drive herd based on location/fitness

        • Alphas – Betas with highest fitness

    • Algorithm ‘seeded’ with individuals evolved with GH, and then optimized with standard GE

    • Typically able to converge to a solution (Santa Fe Ant Trail problem)

Ge crossover
GE Crossover

  • GE crossover found to be ‘destructive’ [O’Neill2003]

    • One-point crossover (standard crossover algorithm)

    • Destroys good trees and generates bloat

  • Exploration of biological-inspired crossovers

    • Homologous

    • Headless-chicken

    • Ripple

Ge crossover1
GE Crossover

  • Homologous

    • History of rules for each grammar stored and aligned

    • Read sequentially and region of similarity noted

    • First crossover points selected as boundary for region of similarity

    • Second from region of dissimilarity

    • Two-point crossover performed

  • Results

    • Standard one- and two-point crossover tend to be more consistent

Ge crossover2
GE Crossover

  • Headless chicken

    • Select fragments for crossover

    • Replaces with randomly-generated bit strings of same length

  • Results

    • Standard one-point crossover performs far better

    • System runs better with crossover switched off

Ge crossover3
GE Crossover

  • Ripple

    • Map codons from middle of parse tree instead of left (preorder traversal) side

    • Find ‘ripple points’

      • One or more sub-trees that can be removed

    • Points on one sub-tree can encode an entirely different sub-tree on another ripple point

  • Results

    • Performs well

    • Tends to search a more global space

Meeting schedule proposal
Meeting Schedule Proposal

  • Proposed update to meeting schedule

    • Move to meeting twice a month, with an email update in the off week

    • Due to limited amount of available weekly development time, this may be a more efficient method to make progress

    • Can schedule interim meetings if discussion / review is necessary in off-weeks

Related work
Related Work

  • Improving Grammatical Evolution in Santa Fe Trail using Novelty Search

    • Urbano and Georgiou, ECAL 2013

  • Swarm Based Population Seeding of Grammatical Evolution

    • Headleand and Teahan, Journal of Computer Science and Systems Biology 2013

  • Crossover in Grammatical Evolution

    • O’Neill, Genetic Programming and Evolvable Machines 2003