Plenary Discussion: What Is Missing in Current Theories of Brain Computation?
summary
The discussion pinpoints concrete gaps: no working algorithm explains how the brain amplifies tiny sensory inputs into rich 3D perception or stores tens of thousands of scenes robustly; there is no accepted mechanistic model for model-based curiosity that drives exploration without external reward; current learning rules fail to explain lifelong consolidation and hidden-unit recruitment for abstraction; theories lack principled ways to compose modules into assembly-based computation or to guide experiments via benchmarks and shared datasets that yield testable predictions.
AI-written summary · about
chapters (20)
- 0:00Brain computation theory gaps
- 1:50Missing principles in models
- 5:12Abstraction barriers in neuroscience
- 8:50Learning rules and modular challenges
- 10:59Multi-scale activity gaps
- 13:26Testable theories needed
- 15:30Population events and assembly issues
- 19:01Brain computation and consciousness
- 20:35Memory capacity challenges
- 26:49Killer experiments with new tools
- 29:03Robust behavior vs training
- 31:20Missing in computational theories
- 35:06Emerging neuroscience paradigms
- 37:39Science vs speculation
- 39:27Theory-experiment cycle
- 41:33Agreements and disagreements
- 45:33Benchmarks in neuroscience
- 51:50Criteria for benchmarks
- 57:36Lessons from successful theories
- 1:01:52Closing remarks
related talks
01h 14mTissue vs silicone musings on the future of deep learning hardware and software
32mHow Neuroscience Makes AI Intelligent
58mStrong Generalization from Small Brains and No Training Data
32mTheoretic analyses of internal model dynamics in the brain
43mCan Cognitive Neuroscience Provide a Theory of Deep Learning...
57mWhat Should a Computational Theory of Cortex Explain?
30mCracking the code: neuronal networks in the brain
01h 13mComputation in Networks of Neurons in the Brain II