The Learning Fly asks whether simulated flies actually learn

Field notes

The index separates fixed connectome wiring from modeled neuron dynamics and synaptic plasticity, then adds validation only when activity or behavior has been compared with real-fly measurements in peer-reviewed work. Its examples include a ViZDoom connection, a Rust neural engine with a Three.js viewer, and FLYT3’s wildfire-damage classifier, where the reported 60.8% result trails a pooled-pixel MLP baseline at 85.7%. The page reports 141 indexed projects, 95 with assigned levels, and four validated against real flies.