Wuhan course in computational neuroscience
Peter Latham
pel@gatsby.ucl.ac.uk
Previous courses:
2021
2022
2023
2024 course

summary of 2022 course Hopefully this will eventually be updated, but don't hold your breath. ;)


Lectures
1. 24.04.15 Recurrent network of randomly connected neurons, part I.
  • zoom recording Passcode: W0k8h4w#
  • youtube
    • 2. 24.04.17 Recurrent network of randomly connected neurons, part II.
  • zoom recording Passcode: cg4Ovv?7
  • youtube
    • 3. 24.04.22 Recurrent network of randomly connected neurons, part III.
  • Very sorry -- I forgot to record the lecture. But almost all of what I said can be found in the second and third lecture from last year's course.
    • 4. 24.04.24 Recurrent network of randomly connected neurons, part IV, plus start of Hopfield networks.
  • zoom recording Passcode: cu23F$NE
  • youtube
    • 5. 24.04.28 Recurrent network of randomly connected neurons, part III.
  • Very sorry -- I forgot to record the lecture. again. But almost all of what I said can be found in the fifth and sixth lecture from last year's course.
    • 6. 24.05.06 Synaptic transmission (Shangbin Chen powerpoint lecture notes)
    7. 24.05.08 Synaptic plasticity (Shangbin Chen powerpoint lecture notes)
    8. 24.05.13 Learning in the brain and in deep networks.
  • zoom recording Passcode: 8+gH^8w2
  • youtube
    • 9. 24.05.15 Learning -- mainly linear regression, but in the overparameterized regime.
  • zoom recording Passcode: ^Sa.f7+x
  • youtube
    • 10. 24.05.20 Backprop in feedforward networks, RNNs and start of transformers.
  • zoom recording Passcode: K?8Gj4s+
  • youtube
    • 11. 24.05.22 A little bit on transformers, but mainly RL.
  • zoom recording Passcode: reqShK3*
  • youtube


    • Notes, papers and problems
    Randomly connected networks:
  • The Wilson-Cowan model, which considered the dynamics of average excitatory and inhibitory firing rates
  • van Vreeswijk and Sompolinsky's classic paper. hard, but thorough
  • firing rate dynamics -- possibly most useful for the construction of nullclines
  • followup experimental paper
  • randomly connected networks A very detailed writeup of randomly connected networks. Probably way too much information

    • Homework problems (to see if you really understand randomly connected networks)
    Problem set 1
  • Problem 4: practice nullclines
    • Problem set 2
  • Problem 1: Wilson-Cowan nullclines
  • Problem 3: sparse connectivity
  • Problem 4: continuous time Hopfield network

    • Attractor networks
  • Hopfield's original paper
  • Notes on the Hopfield model
  • realistic Hopfield networks, simple version
  • realistic Hopfield networks, more complicated version
  • line attractor networks

    • Biologically plausible deep learning
  • one possible method; see the paragraph starting on line 59 for a list of recent approaches

    • Synaptic plasticity
  • Spike-timing dependent plasticity (STDP)
  • How to stabilize STDP
  • BCM rule
  • Ocular dominance columns