Draw a digit — a sparse spiking neural network (5% connection density, LIF neurons, masks from
Drosophila connectome statistics) classifies it with event-driven additions only.
Runs entirely in your browser via ONNX Runtime Web.
· model ·
code
Draw
loading models…
Prediction
draw something, then hit Recognize
The network spikes only ~8% of its neurons per frame; each spike is an addition,
not a multiply-accumulate. Energy estimates use a 45nm CMOS model
(Horowitz 2014: MAC 3.7 pJ vs add 0.9 pJ).
5% densityLIF + surrogate gradient~112× estimated energy saving
Honest boundaries — research artifact, not production. Dense-MLP gap: 1.5 pts (MNIST) / 0.5 pts (Fashion).
Energy figures are analytic (45nm CMOS), not measured on hardware. In our experiments the static
connectome topology gave no advantage — only its statistical laws transferred.
· awareliquid.ai