TinyMo Memory Optimizer

Introduction

TinyMo is a memory optimizing framework for tiny machine learning. It optimizes the peak memory usage of the input model with various methods.

Installation

The prototype implementation requires the following packages:

• tensorflow-cpu
• absl-py
• numpy
• flatbuffers

python3 -m pip install -r requirements.txt

Model Optimization

If you have a TensorFlow Lite model, you can optimize the model using the following command:

python ./optimizer/optimizer.py --model [path to model] --target [target memory usage]

Note that the optimizer will generate the optimzied model in the folder that contains the input model.

Citation

@ARTICLE{chae:2025:tinymo,
  author={Chae, Byungchul and Heo, Seonyeong},
  journal={IEEE Embedded Systems Letters}, 
  title={TinyMo: Graph-Level Memory Optimizer for Tiny Machine Learning}, 
  year={2025},
  volume={17},
  number={3},
  pages={196-199},
  doi={10.1109/LES.2024.3485630}}