Methods in Reinforcement Learning

Artificial Intelligence

• Instincts (predefined logics)
• Machine Learning (statistical data + improvement)
  
  • Supervised Learning (All data)
  • Reinforcement Learning (Target data only)
    • Bruteforce
      • Sequential bruteforce
      • Random bruteforce
    • Intuitive Explore and Exploit (Policy Network)
      • Single action explore and exploit
        • Random action but with known actions in the past
      • Sequence of actions explore and exploit
        • Monte Carlo Tree Search (for max) explore and exploit
    • Explore and Exploit with Bellman (Q-learning)
      • Single action explore and exploit
      • Use q-table (large memory issue, but all cases)
      • Update formula:
        • 
        • learning_rate is for the q-table; the right side of learning_rate is temporal difference
    • Explore and Exploit with Bellman (Q-network)
      • Single action explore and exploit
      • Use q-network (small memory, but similar cases only)
      • Network expected value:
        • 
        • learning_rate is not in this formula, it is defined with optimiser
          
  • Unsupervised Learning (No data)
    • With neuralnet:
      • Autoclustering, etc.
    • Non-neuralnet:
      • K-means, etc.