Fly64
Hooks a MaleCNS model up to Super Mario 64, with a local dashboard and macOS setup. A playful experimental controller; requires your own game ROM.
Neurons are simulated — leaky integrate-and-fire or better — with documented, tuned parameters.
Dispute this rating →The evidence
No evidence, no level.The model is an explicit LIF approximation with stated timestep, membrane time constant, threshold and reset.
https://github.com/ornata/fly/blob/f2f4114e53eaa326e54129f27a5383f93c6957af/fly64/model.py#L23-L29Each step decays the membrane voltage, adds synaptic current plus tonic and noise drive, fires above threshold and resets - voltages change, weights do not.
https://github.com/ornata/fly/blob/f2f4114e53eaa326e54129f27a5383f93c6957af/fly64/model.py#L124-L140The connectome weight matrix is loaded once from disk and thereafter only read.
https://github.com/ornata/fly/blob/f2f4114e53eaa326e54129f27a5383f93c6957af/fly64/model.py#L99-L101
Every neuron is a leaky integrate-and-fire unit with stated parameters (dt 20 ms, tau_m 100 ms, threshold 1.0, reset 0.0) plus tonic drive and seeded noise; membrane voltages decay exponentially, cross threshold, fire and reset each step, and a 250 ms rolling spike-rate window over DNg100 / DNa02-DNg13 / DNp01-DNp10 becomes Mario's stick and jump. That is documented neuron dynamics on measured wiring, so L1. Not L2: the sparse weight matrix is loaded once from the prepared MaleCNS cache and is only ever read when summing input current - a repo-wide search for plasticity, dopamine, STDP, eligibility or any weight assignment turns up nothing outside a test fixture. The README is itself explicit that the wiring is measured and the firing rules are the repo's own.
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.