20fly8
2048 learned in the browser by a FlyWire v783 mushroom-body model: each board cell and tile size drives its own projection neuron (32 of 267), the measured PN -> Kenyon-cell (27,502) and Kenyon-cell -> MBON (62,261) connections carry the smell to 96 output neurons, and approach vs avoidance is read from which dopamine cluster (PAM or PPL1) innervates each MBON. A project-defined dopamine rule driven by a reward-prediction error (TD learning on afterstates) changes only the Kenyon-cell -> MBON gains. Circuit subset; firing-rate model, not spikes. Reports frozen-synapse controls over 500 games: random moves 1,103, naive fly 1,594, trained after 3,000 games 6,986 (512 in 75%, 1024 in 15%), nose rewired 925, synapses shuffled 1,287.
Nobody has read this one closely enough to attach evidence to a level. Until they have, it stays unrated — a guess would be worse than a gap.
Propose a rating →The evidence
No evidence, no level.Nothing filed yet. The description above is carried over from cobanov/awesome-fly and townie/awesome-fruit-fly, which is a source review — not an independent reproduction, and not enough on its own to earn a level.
Record
Nearby
All projects →MaleCNS simulation connected to ViZDoom through modeled visual inputs and a fixed button readout. Includes plasticity experiments and negative validation results; learned survival has not been demonstrated.
Research framework comparing a MaleCNS-constrained sparse recurrent controller with rewired graphs, conventional neural networks, and other controls on simple VizDoom tasks.
Ratings on thelearningfly.com are proposals, not verdicts. Every one of them is arguable in public.