Internal/Teaching AI



Keep Java server, strongly recommend they write players in Python - test the connectivity

Setup network environment to run the server. Can it be r2s server or a virtual machine on the wiki server (talk with Nicholas)?


  • Lab1:
    • reflex agent, using rule-based reasoning engine
      • let's try pyke
      • create 2-3 simple examples
    • simple mobile robot controller
      • let's try Webots
      • we either create our own world, or we use one of the competitions: the rat maze or the savanna world
      • the trick is to have individual decisions be complex enough (we don't want to have too much memory yet)
      • make sure Webots is easy to integrate with pyke!
    • simple poker player
      • demonstrate how to stratify the knowledge (levels of hand quality, evaluate opponents, etc.)
      • how easy is it to calculate winning percentage of any given hand in pyke?
  • Lab 2:
    • search algorighms: A*
    • mobile robot path planning
      • can we make it a car instead of omnidirectional robot?
      • talk to Jennifer/Gaurav about ideas?
      • do it in an abstract manner
      • but maybe we could connect to Webots later for visualisation?
    • poker bidding against known hand & strategy
      • we need to come up with sufficiently complicated strategy!
      • has to have a number of local minima
      • and a number of reasonable heuristics
  • Lab3
    • logic
      • students write knowledge base in FOL
      • the idea is to "do machine learning by hand" - come up with some rules, see which examples are covered, iterate
    • human behaviour analysis and anomaly detection
      • based on data from Jens
      • generated by his simulator
      • KB should explain as many examples as possible
    • poker: deduce opponent's hand
      • based on known cards
      • and description of their strategy
  • Lab 4
    • Bayesian networks
    • healthcare and medical diagnosis
      • match symptoms with causes
      • talk to Anita and Nicholas
    • poker: opponent recognition
      • given non-deterministic strategies, match play history to an opponent
  • Lab 5
    • Learning
      • SVM, neural networks and/or decision trees
    • fingerprint classification (?)
      • talk with Anna (?)
      • maybe face / digit recognition (talk with Stefan K)
    • poker: learn opponent's strategy
      • train on past games
      • try to predict next bids
      • how to deal with hidden information (current hand)?