Fly connectome study suggests path-integration memory lives in synapses, not neural activity
viksit · x · 2026-09-15
An exhaustive pass over every fan-shaped-body population in the MaleCNS (2026) and hemibrain (2020) fly connectomes yields four findings: the long-sought path-integration "sum" is probably stored in synapse strengths at hΔB's outputs rather than sustained loop activity (challenging standard RNN/LLM-style state models); every memory store delivers displacement sign-preserved, with only hΔM and hΔI acting as inverters on the return path; the velocity signals come from cells nobody has named yet (e.g. PS196b feeding ExR2, ExR4, LPsP, FB3A); and the compass has a built-in brake—shifter neurons get 3x more input from the position they write than the one they read—absent from all published navigation models. Each claim is tagged as published, simulated, or hypothesis.
More from Research
- ICML 2025 Position Paper by 16 Researchers: Stop Treating AGI as AI's North-Star Goal — haydenfield · 2026-09-15
- Computer-assisted proof debate: author to release 100GB of independently checkable verifications — DimitrisPapail · 2026-09-15
- Cross-verifying AI math proofs with hundreds of agent calls instead of Lean — DimitrisPapail · 2026-09-15
- Why encoding massive numerical-certificate proofs in Lean is impractical — QuangVDao · 2026-09-15
- Two days doing neuroscience with Fable 5.1: strong research taste, clings to known hypotheses — generativist · 2026-09-15
- Weekly must-read papers: world models and robotic VLA dominate the list — TheTuringPost · 2026-09-15