TDA_KD

Topological Knowledge Distillation for Wearable Sensor Data

Overview

This is re-implementation of the Topological Knowledge Distillation loss described in: E. S. Jeon, H. Choi, A. Shukla, Y. Wang, M. P. Buman, and P. Turaga, “Topological knowledge distillation for wearable sensor data,” in Asilo- mar Conference on Signals, Systems, and Computers, 2022, proceedings Forthcoming.

Requirements

• pytorch>=1.4.0
• python>=3.6.0

Time Series Data Classification

We use time series data classification as an example with a simple architecture. In order to reproduce the results described on the paper, please modify the hyperparameters and model architectures. The users can also change the data to other dataset at their interest. To run the code, please download PAMAP2 dataset as below and create persistence image via PI.ipynb.

Dataset

• Dataset can be downloaded from: http://archive.ics.uci.edu/ml/datasets/pamap2+physical+activity+monitoring
• PAMAP2 preprocessing found in preprocessing_pamap2_2.py

Sample

python3 train_ad.py --epochs 200 --teacher wrn163 --teacher-checkpoint Teaimg/wrn/wrn163_0_image.pth.tar --teacher2 wrn1631 --teacher-checkpoint2 Teasig/wrn/wrn163_0_signal.pth.tar --student wrn1611 --student-checkpoint Teasig/wrn/wrn161_0_signal.pth.tar --cuda 1 --dataset pamap --batch_size 64 --sbj 0 --trial 163m_161_099_id0_00_std --save_weight 0 --seed 1234