---
license: mit
datasets:
- mnist
metrics:
- accuracy: 0.966
- parameters: 1309
pipeline_tag: image-classification
---

A super tiny mnist model to demostrate the potential of the learnable activation - OptAEG-V1.

The model can reach 96.6% accuracy with only 1.3k parameters.

The OptAEG-V1 learnable activation is based on a theory of Arithmetic Expression Geometry which is still in developing.
Please visit the draft papers on [theory](https://github.com/mountain/aeg-paper) and [neural networks](https://github.com/mountain/optim-aeg) for a reference