Can a Fruit Fly Connectome Learn Cube Moves

Creator: @MinLiBuilds

Archived study

Investigates whether a fixed fruit-fly connectome topology can be trained by changing its weights so that cube-move decisions are expressed through identified leg motor neurons, rather than decoded first and translated into leg actions afterward. The archived workflow includes frozen-reservoir comparisons, constrained motor-neuron readouts, and an EfficientCube teacher-distillation workflow using rank-8 LoRA and states at scramble depths 1–9.

Scope

The study evaluates a computational model built from connectome topology. Action classification and complete multistep cube solving are separate evaluation tasks.

Package and provenance

Creator attribution follows the signature on the supplied poster and matches Case Index.csv. The original session export and poster are preserved from AIPOCH-Creator-Use-Cases.zip. This introduction describes the archived research workflow.

Case files

• Session export
• Case poster