My primary research interest lies in selecting and refining training data to enhance the efficiency and alignment of deep learning models. I believe that handling training data is as important as model design in influencing the inductive bias of deep learning models. The appeal of curated data is that it doesn’t vanish and holds the potential for reuse or reprocessing.

Consequently, I have focused on the study of active learning, a leading method to enhance label efficiency through selective data sampling. In my research, Hwang et al., ECCV 2022, I explored data selection techniques to reduce the discrepancy between different domains. Furthermore, in studies like Kim et al., ICCV 2023 and Hwang et al., NeurIPS 2023, I looked into the details of query formulation for label requests in active learning for semantic segmentation.

In addition, our work, Kim et al., IJCV 2022, has investigated the application of multiple-instance learning for weakly supervised learning, aiming to elevate label efficiency.

Beyond label efficiency, I believe in the potential of data selection or refinement in improving model fairness and mitigating data imbalance. As part of this exploration, our study, Kim et al., NeurIPS 2022, was focused on creating strategies to address existing biases within datasets.

Looking ahead, I am interested in enhancing the alignment and efficiency of LLM and MLLM models through data filtering, such as data pruning.

References:

  • Sehyun Hwang, Sohyun Lee, Hoyoung Kim, Minhyeon Oh, Jungseul Ok, and Suha Kwak. Active Learning for Semantic Segmentation with Multi-class Label Query. In NeurIPS, 2023.
  • Hoyoung Kim, Minhyeon Oh, Sehyun Hwang, Suha Kwak, and Jungseul Ok. Adaptive superpixel for active learning in semantic segmentation. In Proc. IEEE/CVF International Conference on Computer Vision (ICCV), 2023.
  • Sehyun Hwang, Sohyun Lee, Sungyeon Kim, Jungseul Ok, and Suha Kwak. Combating label distribution shift for active domain adaptation. In Proc. European Conference on Computer Vision (ECCV), pages 549–566. Springer, 2022.
  • Nayeong Kim, Sehyun Hwang, Sungsoo Ahn, Jaesik Park, and Suha Kwak. Learning debiased classifier with biased committee. In NeurIPS, pages 18403–18415, 2022.
  • Namyup Kim, Sehyun Hwang, and Suha Kwak. Learning to detect semantic boundaries with image-level class labels. International Journal of Computer Vision (IJCV), 130(9):2131–2148, 2022.