Can a fruit fly
predict the future?
Run your time series through the wiring diagram of a real biological nervous system. 166,691 biological neurons. One frozen connectome. Your data.
What happens when you turn an insect nervous system into a computer?
In 2026, researchers published the MaleCNS v1.0 connectome — a complete synapse-resolution electron-microscopy reconstruction of an entire male adult fruit-fly central nervous system, containing over 166,000 neurons and 25.5 million directed synaptic connections spanning the brain, optic lobes, and ventral nerve cord.
FlyCast treats this biological wiring diagram as a fixed recurrent computational reservoir (an Echo State Network). Rather than training billions of artificial weights, the biological wiring remains completely frozen.
When your time-series values arrive, they stimulate 512 annotated sensory neurons. The signal ripples through the actual recurrent synaptic graph, creating rich nonlinear spatial-temporal trajectories.
Only a lightweight linear Ridge readout is trained to map sampled internal network states to your future target values. If the fly predicts better than standard autoregression, the biological wiring provided genuine computational utility. If it fails, the application honestly tells you so.
Chronological data is standardized and deterministically mapped across 512 sensory CNS neurons.
Activity propagates through 25.58M real directed synaptic edges via leaky tanh rate dynamics.
4,096 virtual electrodes sample internal recurrent state vectors across the connectome at each step.
A multi-output Ridge regression learns to project internal states into multi-step future horizons.
Try a benchmark signal through the connectome
Does biological wiring outperform classical models?
Every FlyCast run measures real held-out RMSE against persistence and autoregressive baselines. If the fly wins, you see it. If the fly loses, we display that honestly.