← The index · Desktop flies · by pusulamkendim

FlyWire Neuro

Local FlyWire LIF simulation linked to a persistent 3D body, sensory controls, and recorded runs. Descending-neuron activity selects measured or cached motor behavior. Research prototype.

FlyWireResearch prototype
Fidelity · reviewed
L1 Dynamics

Neurons are simulated — leaky integrate-and-fire or better — with documented, tuned parameters.

Dispute this rating →

The evidence

No evidence, no level.
Rating note

Second reading, downgraded from a proposed L2, on the same ground as two sibling projects running the identical formula. The rule does write into a mutable view of the FlyWire weight values from inside step(), with no optimizer — that part holds. But dW = eta * pre * post * sign − alpha * w applied uniformly across all ~15M synapses is generic pre-post correlation with decay: no dopamine, no compartment, no cell type, nothing that would read differently in a paper about any other animal. L2 asks for a rule grounded in fly biology and this is a general learning rule pointed at a fly graph. The project’s own compare_plasticity.py agrees in effect, reporting that structural wiring utterly dominates the perturbation. L1 is well earned on the FlyWire LIF engine driving a persistent body.

Record

CategoryDesktop flies
DatasetsFlyWire
LevelL1 — reviewed
Listed incobanov/awesome-fly
First indexed2026-09-15

Ratings on thelearningfly.com are proposals, not verdicts. Every one of them is arguable in public.