FLYFEAR
Godot horror prototype where a MaleCNS-derived simulation influences a fly and an external adaptive event policy. Includes training and validation reports.
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 24 discrete leaky-tanh rate iterations over the full connectome, with the author noting it is not a calibrated LIF.
https://github.com/furkancak1r/flyfear/blob/ea04833f681faa049726d099ec4845c4c76b13eb/brain/connectome.py#L145-L147The connectome weight matrix is loaded once from disk and only ever read; no line writes to it.
https://github.com/furkancak1r/flyfear/blob/ea04833f681faa049726d099ec4845c4c76b13eb/brain/connectome.py#L86-L95Reward updates a separate linear-response readout's covariance and target vectors, not any synapse in the fly graph.
https://github.com/furkancak1r/flyfear/blob/ea04833f681faa049726d099ec4845c4c76b13eb/brain/director.py#L105-L117
The full 166,700-neuron / 25.6M-edge MaleCNS graph is simulated as leaky rate units - 24 iterations of state += 0.35 * (tanh(1.5 * W @ state + drive) - state) per decision - with the source comment conceding it is a discrete rate model with no biological clock rather than a calibrated LIF. Simplified leaky units with stated parameters is L1. Not L2: the connectome matrix is loaded from a .npz and appears in exactly one expression, a matrix-vector product, and is never assigned to. All of the learning lives in an external 'director' readout - a discounted ridge-regression bandit over three scare-event actions, updated from player movement or an optional 1/2/3 rating - which sits outside the fly circuit entirely. The README states the point plainly: what is learned is scare-event preference, while the flight control rule is fixed.
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.