VolumE Sampling for Streaming Active Learning (VeSSAL)

This repository is an implementation of the VeSSAL batch active learning algorithm. Details are in the following paper:

A. Saran, S. Yousefi, A. Krishnamurthy, J. Langford, J.T. Ash. Streaming Active Learning with Deep Neural Networks. ICML 2023.

For a quick overview of the approach, please check the talk and poster presented at ICML 2023.

This code was built on Kuan-Hao Huang's deep active learning repository, and Batch Active learning by Diverse Gradient Embeddings.

Dependencies

To run this code fully, you'll need PyTorch and Torchvision and scikit-learn. We've tested our code with PyTorch 1.8.0, 1.13.0 and Python 3.7, 3.8.

Running an experiment

python run.py --model resnet --nQuery 1000 --data CIFAR10 --alg vessal
runs an active learning experiment using a ResNet and CIFAR-10 data, querying batches of 1,000 samples according to the VeSSAL algorithm. This code allows you to also run each of the baseline algorithms used in our paper.

python run.py --model mlp --nQuery 10000 --data SVHN --alg conf
runs an active learning experiment using an MLP and SVHN data, querying batches of 10,000 with confidence sampling.

Bibliography

If you find our work to be useful in your research, please cite:

@article{saran2023streaming,
  title={Streaming Active Learning with Deep Neural Networks},
  author={Saran, Akanksha and Yousefi, Safoora and Krishnamurthy, Akshay
  and Langford, John and Ash, Jordan T.},
  journal={arXiv preprint arXiv:2303.02535},
  year={2023}
}