Fruit Fly Simulation (Xenova)
Joshua Lochner’s WebGPU MaleCNS demo: paint / stimulate 166,700 neurons and watch an articulated Three.js fly. Connectivity is biological; movements are programmed. Verified Hugging Face Space (not a GitHub repo).
Neurons are simulated — leaky integrate-and-fire or better — with documented, tuned parameters.
Dispute this rating →The evidence
No evidence, no level.The LIF parameters (rest, threshold, tau_m, tau_s, refractory, dt) are stated in the simulation source.
https://huggingface.co/spaces/Xenova/fruit-fly-simulation/blob/776d115ee5aa934578a87fd6d260d138084f59c1/src/brain.js#L1-L10The integration step evolves membrane voltage exponentially toward rest, thresholds, and resets; synaptic weights are only read.
https://huggingface.co/spaces/Xenova/fruit-fly-simulation/blob/776d115ee5aa934578a87fd6d260d138084f59c1/src/brain.js#L82-L92The authors state the model is a leaky integrate-and-fire adapted from Shiu et al. and that the body movements are engineered, not validated.
https://huggingface.co/spaces/Xenova/fruit-fly-simulation/blob/776d115ee5aa934578a87fd6d260d138084f59c1/README.md#L58-L62
166,700 MaleCNS neurons and 25.6M connections driven by a leaky integrate-and-fire model adapted from Shiu et al., with stated rest, threshold (-45 mV), tau_m (20 ms), tau_s (5 ms) and refractory parameters and exponential-Euler integration, running in WebGPU with a JavaScript fallback. Not L2: there is no training of any kind and no weight is ever written; the connectome weights are downloaded as verified fixed chunks, and the only user action is painting stimulation onto neurons. The fly's walking, turning and flight are crafted animations driven by computed firing rates through engineered gains and thresholds, which the README labels as illustrative and not validated predictions of fly behaviour; motion is not learning, and body motion does not even feed back into the network.
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.