Computation in Networks of Neurons in the Brain II
summary
Argues that diverse neuron and synapse properties — documented in Allen Institute data and summarized by the three-parameter UDF model for short‑term plasticity — break simple rate‑based and sign‑flip circuit implementations (many excitatory→excitatory synapses are strongly depressing while excitatory→inhibitory are facilitating). Cites simultaneous recordings (dozens to hundreds; an 87‑cell example) showing continuous assembly trajectories and place‑code drift over days, reviews liquid computing and FORCE readout training, and proposes stochastic synaptic plasticity with a temperature that makes learning a continual sampling process.
AI-written summary · about
chapters (17)
- 0:00Intro
- 1:35Diversity
- 2:30Shortterm plasticity
- 5:20No computation
- 11:27No network states
- 13:09No dynamical system
- 15:50No diversity of units
- 18:22Liquid no computing
- 20:54Learning algorithms
- 22:26Multiplying neural codes
- 23:11Typical network activity
- 26:49No assembly coding
- 28:06Synaptic plasticity
- 39:19Open question
- 48:17Stochastic nature of the network
- 50:02No lifelong learning
- 53:07In principle
related talks
47mDynamic Neural Network Structures Through Stochastic Rewiring
48mControl of Synaptic Plasticity in Deep Cortical Networks
51mThe Brain as an Efficient, Robust Adaptive Learner
31mNeuromorphic computing in spiking neuron-astrocyte networks
27mWhat can artificial neural networks learn from biological neuromodulatory systems?
47mCapacity of Neural Networks for Lifelong Learning of Composable Tasks
12mComputation with sequences of neuronal assemblies
43mCoin-Flipping in the Brain: Probabilistic Computation and Learning in the Assembly Model