jump n bump ai
Download
Skip this Video
Download Presentation
Jump’n’Bump AI

Loading in 2 Seconds...

play fullscreen
1 / 7

Jump’n’Bump AI - PowerPoint PPT Presentation


  • 90 Views
  • Uploaded on

Jump’n’Bump AI. The Search For Adequacy. Jump’n’Bump Refresher. ?. Progress. First try: Evolving weights over fixed topology networks. Result: Best results when evolved against stationary target (unimpressive). Problem: Search space too large?. Evolving with NEAT.

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

PowerPoint Slideshow about ' Jump’n’Bump AI' - deirdre-holman


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
jump n bump ai

Jump’n’Bump AI

The Search For Adequacy

progress
Progress
  • First try: Evolving weights over fixed topology networks. Result: Best results when evolved against stationary target (unimpressive).
  • Problem: Search space too large?
evolving with neat
Evolving with NEAT
  • Java implementation: ANJI (Another Neat…)
  • Many different evolution opportunities:
    • Direct Tournament (play against fixed opponents)
    • K-Random Opponents
    • Single-Elimination Tournament
  • Best results seem to be from direct tournament against hand-coded AI.
  • But need quantitative answer, not qualitative observations
the grand tournament
The Grand Tournament
  • 13 Contestants
    • Fixed Topology Networks
    • NEAT Evolved Networks
      • Direct, single-elimination, and k-random opponents tournaments
    • Hand Coded AI
  • 98000 Jump’n’Bump Match Round-Robin
  • Conclusion: Jump’n’Bump playing ability is not transitive.
  • Overall Winner: Hand Coded AI
conclusions
Conclusions
  • NEAT seems to be able to evolve an AI better than anything you can pit it against…
  • …but to be better than human, you would already need a human-class AI
  • Also: AIs that are evolved against each other generalize poorly
  • Maybe the Internet could be a solution
ad