L2D in Time Series

Extending the learning-to-defer (L2D) framework to time series data
Inovating a novel architecture.

Adversarially Robust Learning-to-Defer for Classification and Regression

Extend the robustness framework of Montreuil et al. (2025a) to one-stage scenario, where considering the interaction between learning the predictor and adapting the experts.
Code Base: Currently Not Public
Collaborator & Supervisor: Yannis Montreuil [NUS & CNRS]
arXiv Preprint: https://arxiv.org/abs/2510.10988

Road Sign Classification with Denoising Pipeline Approach

Real-world images are affected by image noise, which degrades the quality and reduces reliability of traffic sign recognition. This research proposes a denoiser-classifier pipeline approach, which uses a pre-trained color image denoiser, and compares it against an end-to-end approach. We evaluated CNN and ResNet-18 on GTSRB under Gaussian noise.
Code Base: classify_denoise_SwinIR.
Collaborators: Yuqi Sun [UMich], Jiaqi Zhang [McGill], Huayu Liu [UC Irvine], Zhuoyang Hu
Supervisor: Professor David Woodruff [CMU]