Project on designing stealthy backdoor triggers in CLIP-style models by aligning visual and textual embeddings (Liang et al., CVPR 2024).

Project on crafting targeted perturbations for multi-label classifiers that respect semantic label dependencies (Mahmood et al., CVPR 2024).

Exploration of ST++ techniques—strong data augmentations and selective re-training—to boost segmentation with limited labels (Yang et al., CVPR 2022).

Advanced scene representation using multi-branch hash tables and NeRF-style encoding for large-scale dynamic urban reconstructions (Turki et al., CVPR 2023).

Implementation of a Bi-GRU with self-attention framework for classifying emotions from audio/text (Wu et al., ICASSP 2023).

AutoGAN (Parti) (Google Research, TMLR 2022)

Control GAN (B. Li et. al, NIPS 2019).

Imagen (Google Research, NIPS 2022).

Study on automatically prompting SAM with an RL agent for diverse segmentation tasks (Huang et al., CVPR 2024).

Integrating external APIs into reward models for more accurate, transparent decision processes (Li et al., ICLR 2024).

Development of a second-order functional connectivity embedding plus domain-adaptation to improve autism detection across diverse sites (Kunda et al., IEEE TMI 2022).

Reviewing and improving the Mutual Correction Framework (MCF) for refining medical image segmentation masks via mutual corrections solving confirmation bias (Wang et al., CVPR 2023).

Investigated the gap between declared training context length and empirical “effective” context length, analyzed positional-encoding biases and shifting method (Chenxin, ICLR 2025).