Emergent elasticity in the neural code for space

on YouTube ↗54m

" summary

Evidence shows grid code coherence decays within 40 seconds without landmarks, so a solvable attractor-plus-Hebbian model is built: landmark synapses learn by plasticity and behave like particles on springs — an emergent elasticity whose rest lengths match physical landmark separations. The model predicts small history-dependent shifts (~1–2 cm) in grid fields conditioned on last border touch and a bifurcation under virtual-reality gain changes producing either steady phase lags or full remapping; large datasets and VR recordings confirm these signatures.

AI-written summary · about

" chapters (29)

1. 0:00Intro
2. 2:18The extraction of spatial position from the periphery
3. 3:25A fundamental consistency requirement relating path integration to landmark based error correction
4. 4:17Talk outline
5. 9:24Options have different dependencies on trajectory angle
6. 10:54Error is corrected in a direction- dependent manner
7. 11:531D Attractor Model
8. 14:14Landmark influence
9. 15:05Landmark Learning
10. 15:55Don't worry (1): there are equations underlying all this Don't worry (2): I won't show them again
11. 21:33Simplest Case
12. 21:46Before Learning
13. 22:57After Learning
14. 23:36Interpretation
15. 25:37Emergent Mechanical Relaxation of Internal Maps Particles - Landmark Synapses
16. 27:33Generalization to two dimensions
17. 29:09Distortions
18. 33:26Matches Experimental Observations
19. 34:17Bending of Firing Fields
20. 35:47Topological Defects
21. 36:48How to create a topological defect in grid cells
22. 38:38Head-fixed virtual reality offers near-perfect cue control
23. 39:17Cell types in VR give analogous behavior
24. 40:31Gain manipulations reveal influence of visual and locomotor speed cues on grid and border cells
25. 41:19Attractor-Landmark Model
26. 45:07Gain Increase: Attractor is slower than landmarks
27. 45:21Weak Gain Increase
28. 45:48Strong Gain Decrease
29. 46:05Landmark Friction Why doesn't field number double?

" related talks

• 30mCracking the code: neuronal networks in the brainSabria Lagoun · WeAreDevelopers AI Congress Vienna 2018
• 01h 13mComputation in Networks of Neurons in the Brain IIWolfgang Dr. Maass (University of St. Gallen) · The Brain and Computation Boot Camp
• 58mStrong Generalization from Small Brains and No Training DataBruno Olshausen · Unknown Futures of Generalization 2024
• 01h 22mSpace is a latent sequence: A theory of hippocampus and PFCDileep George · Understanding Higher-Level Intelligence from AI, Psychology, and Neuroscience Perspectives
• 31mNeuromorphic computing in spiking neuron-astrocyte networksSusanna Gordleeva · DSC Europe 24
• 36mScaling Laws, Emergent Behaviors, and AI DemocratizationIrina Rish, PhD · ODSC East 2023
• 13mEnhanced EEG Data Classification with 3D SNN and On-Line Learning ESNDimitar Penkov · DSC Europe 23
• 12mComputation with sequences of neuronal assembliesMax Dabagia · Understanding Higher-Level Intelligence from AI, Psychology, and Neuroscience Perspectives