ProtoHAR: Prototype Guided Personalized Federated Learning for Human Activity Recognition (IEEE JBHI 2023)

Dongzhou Cheng

Abstract (paper)

Benchmark

We offer a benchmark for USC-HAD and HARBOX.

1. git clone the repo

git clone https://github.com/cheng-haha/ProtoHAR.git

2. Enter the current folder

cd {yourfolder}

3. Generate heterogeneous data sets

NOTE:args.dataset_dir = {your dataset path}

python  data/uschad/uschad_subdata.py
python  data/harbox/harbox_subdata.py

4. Usage

bash runexp.sh

Run details

1. USC-HAD has 14 clients, HARBOX has 120 clients
2. the learning rate of USC is 0.001, HARBOX:0.01