- Category
- AI Tools
- Type
- TOOL
- Date
About
Sparse-SNN is a browser-runnable demo of a sparsely connected spiking neural network (5% connection density) built from leaky-integrate-and-fire neurons, with connectivity masks derived from Drosophila connectome statistics. It classifies hand-drawn MNIST and Fashion-MNIST digits via ONNX Runtime Web, activating only about 8% of neurons per frame and estimating roughly 112x lower energy use than a dense model by replacing multiply-accumulates with simple additions.
Why it made the leaderboard
You can draw a digit and watch a 5%-density spiking network (Drosophila connectome-inspired) classify it in the browser using only additions, with the author's own honest accounting of where the approach underperforms a dense baseline.
Tags
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.