flybrain-snake
Live MaleCNS (~165k neurons) playing Snake, with the control mapping documented instead of hidden.
Neurons are simulated — leaky integrate-and-fire or better — with documented, tuned parameters.
Dispute this rating →The evidence
No evidence, no level.The neuron model is LIF with explicitly stated membrane, synaptic, adaptation and refractory constants.
https://github.com/charbelkassab/flybrain-snake/blob/6dcaacb62f2efdc3a624be06962084052664f373/brain.py#L20-L31The integration step reads the connectome matrix W to inject current and never writes to it; only voltage, conductance and adaptation state change.
https://github.com/charbelkassab/flybrain-snake/blob/6dcaacb62f2efdc3a624be06962084052664f373/brain.py#L104-L122The only trained component is a separate multinomial logistic decoder fitted by gradient descent on descending-neuron spike counts.
https://github.com/charbelkassab/flybrain-snake/blob/6dcaacb62f2efdc3a624be06962084052664f373/experiment.py#L50-L73
Whole-CNS leaky integrate-and-fire over the 165k-neuron MaleCNS graph with fully stated parameters (V_rest -52 mV, V_th -45 mV, tau_m 20 ms, tau_syn 5 ms, 2 ms refractory, spike-frequency adaptation), following Shiu et al. with documented deviations. Not L2: the connectome matrix W is loaded from disk and only ever read (`self.W[:, delayed].sum`); the only thing fitted is a separate multinomial logistic decoder over the 1,314 descending neurons, trained by gradient descent in experiment.py. The README is explicit that 'the brain itself is not trained', and the project's best result (mean 25.4) comes from that bolt-on readout. Its scrambled-wiring control is unusually good practice but is an ablation of fixed wiring, not plasticity.
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.