Coin-Flipping in the Brain: Probabilistic Computation and Learning in the Assembly Model
Talkat Understanding Higher-Level Intelligence from AI, Psychology, and Neuroscience Perspectives · 2024-06-24..06-28·
summary
Model uses a directed Erdős–Rényi connectome with constant brain areas of n neurons where k out of n fire (k ≈ √n) via winner‑take‑all inhibition, Hebbian plasticity with homeostatic renormalization, and only local synaptic updates. Assemblies form and stabilize in about a dozen presentations, show pattern completion from ~10% cues, and sequence presentations learn finite‑state machines and Markov chains (proven for out‑degree‑2) from data. Ambient noise modeled as Gaussian perturbation yields probabilistic choices that converge in log k steps; claims have matching simulations and theorems, but stochastic tasks require ~10–20× larger networks.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 2:17Brain as a computational model
- 5:51The mathematical brain model
- 11:00Assembly emergence
- 17:25Memory and pattern association
- 19:03Learning through sequences
- 21:24Finite state automata learning
- 24:26Statistical learning
- 29:25Generative model capabilities
- 30:47Hierarchical structures
- 34:06Conclusion and Q&A
related talks
55mComputation with Assemblies
47mCapacity of Neural Networks for Lifelong Learning of Composable Tasks
01h 13mComputation in Networks of Neurons in the Brain II
49mAn Algorithmic Theory of Brain Networks
47mDynamic Neural Network Structures Through Stochastic Rewiring
58mStrong Generalization from Small Brains and No Training Data
01h 07mTheory of Computation II
51mThe Brain as an Efficient, Robust Adaptive Learner