fly-craftax
Connectome simulation connected to Craftax, with PPO training of a descending-neuron readout, baseline comparisons, and a replay viewer. Research prototype.
Neurons are simulated — leaky integrate-and-fire or better — with documented, tuned parameters.
Dispute this rating →The evidence
No evidence, no level.Documented Shiu LIF parameters for the connectome simulation.
https://github.com/liuzihe02/fly-craftax/blob/2fe4145b47a30c3daf45447b17832b86dc345dfc/flycraftax/brain.py#L12-L22The per-timestep membrane update, threshold test and reset.
https://github.com/liuzihe02/fly-craftax/blob/2fe4145b47a30c3daf45447b17832b86dc345dfc/flycraftax/brain.py#L109-L118PPO's trainable parameters are only the linear readout and value head; the module docstring states the brain stays frozen.
https://github.com/liuzihe02/fly-craftax/blob/2fe4145b47a30c3daf45447b17832b86dc345dfc/flycraftax/ppo.py#L1-L14
A Shiu et al. 2024 LIF reimplemented in JAX with full parameters (dt 0.1 ms, v_rest/v_reset -52 mV, v_th -45 mV, t_mbr 20 ms, tau 5 ms, refractory 2.2 ms, delay 1.8 ms, w_syn 0.275 mV) and careful attention to matching Brian2's refractory write semantics. That earns L1. Not L2: the project's own PPO module opens with the line that the brain stays frozen, and the parameter dict handed to the optimiser contains only the linear readout weights W, b and the value head vw, vb over descending-neuron rates. No connectome weight is ever in the optimiser or written during a rollout. This is the trained-readout-on-a-frozen-graph pattern, done cleanly and labelled as such.
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.