Please enable JavaScript.

Coggle requires JavaScript to display documents.

Artificial Intelligence (Concepts (Agents (Perceive the environment…

- - - - Advantage: domain-knowledge easy to express
      - Disadvantage: cannot deal with situations not anticipated by programmer
      - Popular to use behaviour trees; very fast algorithm, but not sure how far reaching the tree should be
    - - Unsupervised (Reinforcement learning)
      - Supervised (Classification)
      - Evolutionary: from a pool of possible controllers, try them out, select the ones that do best, and mutate and recombine for a number of iterations, keeping the best
      - Advantage: does not require much knowledge in principle
      - Disadvantage: in practice, hard to know which features to learn, and is slow to learn
    - - Some new problem -> describe problem -> use an algorithm -> solution competitive with huamn-made specialised problem
      - Specify model for problem (action, initial situation, goals and sensors)
      - Let a solver compute the controller automatically
      - Actions, sensors, goals -> SOLVER -> CONTROLLER -> (actions) World <- (observations to the controller)
      - Advantages: powerful, quick, flexible & clear, intelligent & domain-independent
      - Disadvantage: need a model, computationally intractable, efficiency loss without any domain-specific knowledge
      - Trade-off between 'automatic and general' vs 'manual work but effective'