Ruby Project
NeoForge Minecraft 1.21.1 mod that gives a mob a FlyWire-derived LIF brain (photoreceptors, descending neurons, Kenyon cells, MBON, DAN) with dopamine-gated KC→MBON learning. Real wiring plus a few synthetic Minecraft entry neurons; the author documents what is fabricated.
Synapses change under a modeled rule grounded in fly biology — for example dopamine-gated change on Kenyon-cell → MBON synapses in the mushroom body. Learning begins here.
Dispute this rating →The evidence
No evidence, no level.The dopamine-gated update writes a per-synapse delta indexed by a real connectome edge.
https://github.com/wavendis/ruby-project/blob/693d97608d542209cd3cae8c0349f84e244b4ba5/src/main/java/net/wavendis/flybrain/brain/BrainEngine.java#L112-L127The LIF propagation uses the connectome weight plus the learned delta, so plasticity changes the circuit that runs.
https://github.com/wavendis/ruby-project/blob/693d97608d542209cd3cae8c0349f84e244b4ba5/src/main/java/net/wavendis/flybrain/brain/BrainEngine.java#L58-L66Reinforcement is applied every tick during a reward, over KC->MBON pairs, gated by the real reward DANs.
https://github.com/wavendis/ruby-project/blob/693d97608d542209cd3cae8c0349f84e244b4ba5/src/main/java/net/wavendis/flybrain/entity/FlyBrainEntity.java#L279-L283The plastic pairs are enumerated from connectome synapses whose pre is a Kenyon cell and post an MBON, by FlyWire role.
https://github.com/wavendis/ruby-project/blob/693d97608d542209cd3cae8c0349f84e244b4ba5/src/main/java/net/wavendis/flybrain/entity/FlyBrainEntity.java#L528-L534
Three-factor plasticity on real FlyWire synapses, verified to the assignment. `threeFactorReinforce` computes delta = learningRate * pre * post * mean(DAN output) and accumulates it into `weightDelta[synapseIdx]`, where the index comes from `connectome.synapseIndexFor(pre, post)` — an edge of the loaded FlyWire graph, not a bolt-on layer. The LIF step then propagates through `effectiveWeight = s.weight() + weightDelta[i]`, so the learned change is what the circuit actually runs on. It is called every tick during a reward, over the KC->MBON pairs enumerated from the connectome by role, with dopamine read from the real `dan_reward` neurons' own output. The shipped graph is real: 13,359 FlyWire root IDs and 578,889 synapses, including 500 Kenyon cells, 75 MBONs and 307 reward DANs; only 5 synthetic entry neurons (the smell and ground-contact bridges) are fabricated, and the README says so. No validation mark: no published comparison against fly experiments, and the README states plainly that every tuning constant including REWARD_LEARNING_RATE is hand-tuned with no biological basis for its value.
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.