In semester two of 2019 we ran a seminar on reasoning in the context of deep reinforcement learning with an aim to understand AlphaGo and related breakthroughs, such as AlphaStar. Along the way we looked at deep learning more generally. Some relevant background information:

There are three main components of AlphaGo: Monte-Carlo tree search, deep learning and reinforcement learning, and we will have talks on all three aspects. One important running theme will be the dichotomy between problems with small and large state spaces, and the corresponding need for function approximation (the successful incorporation of which is what makes AlphaGo scientifically interesting).

Talk schedule:

  • Lecture 0: Geoff Hinton video “Deep learning” (brief notes by DM)
  • Lecture 1: Daniel Murfet “Introduction to reinforcement learning” (notes, video)
  • Lecture 2: James Clift “Turing and Intelligent Machinery” (notes, video, Turing’s paper)
  • Lecture 3: Thomas Quella “Hopfield networks and statistical mechanics” (notes, video)
  • Lecture 4: Will Troiani “Universal approximation by feedforward networks” (notes, paper)
  • Lecture 5: Susan Wei “Deep learning as a continuous dynamical system” (video, paper)
  • Lecture 6: Mingming Gong “Convolutional neural networks”
  • Lecture 7: Daniel Murfet “AlphaGo” (notes, video)