AxonWeave
Python library exposing MaleCNS as a sparse, trainable substrate for NumPy, PyTorch, and TensorFlow while keeping source topology and modeling policies explicit.
Neurons are simulated — leaky integrate-and-fire or better — with documented, tuned parameters.
Dispute this rating →The evidence
No evidence, no level.The STDP rule writes its updated values back into the connectome CSR's weight data.
https://github.com/dhakalnirajan/axonweave/blob/622eaf5ea1f2e6a7dce0fa417e5424fbe962fc6d/python/axonweave/learning/__init__.py#L32-L62The plasticity rule is fully implemented: exponential pre/post traces, LTP minus LTD, reward as the third factor, optional weight bounds.
https://github.com/dhakalnirajan/axonweave/blob/622eaf5ea1f2e6a7dce0fa417e5424fbe962fc6d/python/axonweave/native.py#L178-L200The agent loop steps the dynamics on the plastically modified connectome weights and updates them from the environment's reward each step.
https://github.com/dhakalnirajan/axonweave/blob/622eaf5ea1f2e6a7dce0fa417e5424fbe962fc6d/python/axonweave/experiment/__init__.py#L81-L124A test asserts the connectome working weights actually change after running with the STDP rule.
https://github.com/dhakalnirajan/axonweave/blob/622eaf5ea1f2e6a7dce0fa417e5424fbe962fc6d/tests/test_experiment.py#L114-L120The authors state the plasticity rules are biology-inspired computational rules and make no claim that Drosophila uses them at these synapses.
https://github.com/dhakalnirajan/axonweave/blob/622eaf5ea1f2e6a7dce0fa417e5424fbe962fc6d/docs-site/content/limitations.md#L60-L69
Second reading, downgraded from a proposed L2. The plasticity is real and it does write into the connectome CSR during the run — that half is not in dispute. What fails is the other half of L2: the rule must be grounded in fly biology, and this one is a generic three-factor STDP applied uniformly to every MaleCNS edge, where the "dopamine" term is whatever scalar the environment happens to return. There is no compartment, no mushroom body, no cell type. The authors say it themselves in their limitations, writing that no claim is made that Drosophila uses these rules at these synapses. A general-purpose plasticity library pointed at fly-shaped data is L1: if generic STDP on a connectome counted as L2, the level would stop meaning anything. L1 stands on the LIF and adaptive-LIF dynamics over synapse-count-derived MaleCNS weights.
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.