language:
- en
task_categories:
- time-series-forecasting
tags:
- time-series
- self-supervised-learning
- representation-learning
- time-series-classification
- time-series-regression
This dataset repository contains the preprocessed data used in our NeurIPS 2025 paper Learning Without Augmenting: Unsupervised Time Series Representation Learning via Frame Projections.
Paper: Learning Without Augmenting: Unsupervised Time Series Representation Learning via Frame Projections Project Page: https://neurips.cc/virtual/2025/poster/118514 GitHub Repository: https://github.com/eth-siplab/Learning-with-FrameProjections
Datasets Included
This repository includes all nine datasets across five time-series tasks in different ready-to-use formats, as used in the paper:
- Heart rate estimation: IEEE SPC12, IEEE SPC22, DaLiA
- Activity recognition: HHAR, USC
- Cardiovascular disease classification: CPSC2018, Chapman
- Step counting: Clemson
- Sleep staging: Sleep-EDF
Sample Usage
This dataset contains the preprocessed data that can be used with the associated code from the Learning-with-FrameProjections GitHub repository. Here are the quickstart commands for pre-training and testing:
Pre-training + testing (our method)
python main.py \
--framework isoalign \
--backbone resnet \
--dataset ieee_small \
--n_epoch 256 \
--batch_size 1024 \
--lr 1e-3 \
--lr_cls 0.03 \
--cuda 0 \
--cases subject_large
Supervised baseline
python main_supervised_baseline.py \
--dataset ieee_small \
--backbone resnet \
--block 8 \
--lr 5e-4 \
--n_epoch 999 \
--cuda 0