FlyBrain
Browser-based FlyWire FAFB v783 LIF simulation with food, touch, light, and temperature inputs and a live neural activity display.
Neurons are simulated — leaky integrate-and-fire or better — with documented, tuned parameters.
Dispute this rating →The evidence
No evidence, no level.The simulation step is leak, threshold, spike, reset and refractory - a documented integrate-and-fire neuron model over the connectome CSR.
https://github.com/snedea/flybrain/blob/9191824d17871b7851645782d53d23f213ddb938/js/sim-worker.js#L262-L327Edge weights are read from the binary connectome and normalized once at load; the step loop only reads values[j].
https://github.com/snedea/flybrain/blob/9191824d17871b7851645782d53d23f213ddb938/js/sim-worker.js#L120-L136'reward-gated plasticity' is a comment on a fixed hand-set weight of 3 in the coarse group graph, with no accompanying update rule.
https://github.com/snedea/flybrain/blob/9191824d17871b7851645782d53d23f213ddb938/js/constants.js#L222-L235
The 139,255-neuron FlyWire FAFB v783 graph runs in a Web Worker as an integrate-and-fire model with a leak rate, a spike threshold of 1.0 and a 3-tick refractory period, which clears the L1 bar. Not L2: the CSR edge values are filled and normalized once when the binary connectome loads (and re-ordered once for group-sorted access), then only ever read inside the propagation loop. The strings 'reward-gated plasticity' and 'punishment-gated plasticity' in the source are comments sitting next to fixed, hand-authored edge weights in a coarse ~70-group graph - there is no update rule behind them. Worth flagging for the index: the fly's actual behaviour is driven by that hand-authored group graph inherited from the worm-sim fork, not by the 139K connectome sim, which drives the activity display.
Record
Nearby
All projects →Ratings on thelearningfly.com are proposals, not verdicts. Every one of them is arguable in public.