OptuMNIST

75% to 85% accuracy on MNIST with 355 parameters. Can you do better?

If so, contribute!

What is this?

We aim to better understand the Pareto front of accuracy and number of parameters for MNIST.

This may prove useful to help understand hyperparameter tuning for frugal models.

Do you have ideas for improvements?

How to use

With TorchUncertainty

Install torch, then TorchUncertainty with

pip install torch-uncertainty

Then run the model with

python optumnist/optumnist-v1.py

The dataset will be downloaded automatically.

Without

Install torch, build your own trainer and get the model and optimization procedure from optumnist/optumnist-v1.py.

Reference

OptuMNIST-v1 has been found with Optuna:

Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, Masanori Koyama. Optuna: A Next-generation Hyperparameter Optimization Framework. In KDD 2019.