Quantization of Convolutional Neural Networks in HAR

This repository contains experiments and a framework for quantization of neural networks with PyTorch. Most results can be found in this notebook.

Directories

• The common directory contains common code used by many experiments
• The train_scripts directory contains all experiments. They only require a Sacred MongoDB for metric logging.
• The eval_notebook directory contains the analysis of all logged metrics produced by the experiments.

Thesis

All experiments were part of my bachelor thesis Quantisierung von Convolutional Neural Networks für die Aktivitätserkennung [German, PDF].