FLYNN
FlyWire FAFB v783 as a ~139k-unit sparse RNN trained with DAgger to drive a wheeled MuJoCo robot, vs small-world and CNN baselines. Wang & Chen, arXiv:2607.00025.
Neurons are simulated — leaky integrate-and-fire or better — with documented, tuned parameters.
Dispute this rating →The evidence
No evidence, no level.Unit dynamics are an explicit leaky rate update with a per-cell-type leak rate, which is what earns L1.
https://github.com/ben-gitdev/fly-gym/blob/50a8e8f51bbd184fe2f55c30a9fdc8dca9891864/models/connectome_rnn_model.py#L198-L203The connectome weight vector is a trainable nn.Parameter gated by a train_rnn_weights flag, which shared_config sets to True.
https://github.com/ben-gitdev/fly-gym/blob/50a8e8f51bbd184fe2f55c30a9fdc8dca9891864/models/connectome_rnn_model.py#L118What changes the connectome weights is a torch.optim.Adam optimizer, not a biological rule.
https://github.com/ben-gitdev/fly-gym/blob/50a8e8f51bbd184fe2f55c30a9fdc8dca9891864/shared_config.py#L161-L166Weight changes happen in an offline DAgger loop that backpropagates an imitation loss and steps the optimizer, then saves a fixed checkpoint - nothing changes during an evaluation rollout.
https://github.com/ben-gitdev/fly-gym/blob/50a8e8f51bbd184fe2f55c30a9fdc8dca9891864/train_connectome_rnn_dagger.py#L940-L960
Every unit is one FlyWire FAFB v783 neuron and every recurrent weight one synapse-count-weighted connection, run as a leaky rate unit h_new = (1-alpha)*h + alpha*phi(W@h + b) with a per-cell-type leak rate, plus an explicit virtual-retina front end. That documented unit dynamics earns L1. Not L2 even though the connectome weights themselves are trainable: the change is driven by torch.optim.Adam minimising a DAgger imitation loss against a VFH*+PID planner, not by any modelled neuromodulator or fly learning rule, and it happens in an offline training script that saves a checkpoint - at rollout time the weights are fixed. Gradient descent on a connectome-shaped weight matrix is fitting, not plasticity. Note also that the connectome edge list and checkpoints are not in the repo (too large, no hosted download yet), so the wiring itself cannot be verified from a fresh clone.
Record
Nearby
All projects →Research code for the Shiu et al. connectome-based leaky integrate-and-fire model, including activation/silencing experiments, notebooks, and FlyWire data configuration.
FlyWire whole-brain LIF implementation based on Shiu et al., with multiple simulation backends and benchmarking tools. This repository supplies the neural model; it is not a complete embodied demo package.
Ratings on thelearningfly.com are proposals, not verdicts. Every one of them is arguable in public.