Spaces:
Running
Running
| title: Neuron Runner | |
| emoji: 🧠 | |
| colorFrom: blue | |
| colorTo: purple | |
| sdk: static | |
| pinned: false | |
| license: mit | |
| short_description: RL neural net learns a platformer in real time | |
| # Neuron Runner | |
| **By [cahlen](https://huggingface.co/cahlen)** | [GitHub](https://github.com/cahlen/neuron-runner) | |
| A neural network learns to play a 2D platformer in real time, entirely in the browser. | |
| - **CoFrNet** (default) — continued-fraction network ([Puri et al., NeurIPS 2021](https://arxiv.org/html/2506.05586)): parallel ladders of nested fractions with a dedicated ladder-rung visualization and feature-attribution panel | |
| - **MLP** — classic two-layer perceptron with live node-and-edge activation visualization | |
| - **REINFORCE** policy-gradient training, running live via TensorFlow.js (WebGPU / CPU) | |
| - Three.js Mario-themed game renderer | |