About me
I'm a CS PhD student at the University of Texas at Dallas, where I study on computer vision and computer audition following Prof. Yapeng Tian.
Prior to UTD, I got a Bachelor of Science in Mathematics from Sichuan University.
News
11/2024: I was in the top reviewers for NeurIPS 2024!
09/2024: I will serve as a Reviewer for ICLR 2025.
08/2024: One paper is accepted by UCM-KDD 2024. Congratulations to Kevin!
07/2024: I will serve as a Reviewer for NeurIPS 2024.
06/2024: I will join Prof.Prof. Yapeng Tian's lab to pursue a PhD degree at University of Texas at Dallas! 🌱
05/2024: One paper is accepted by CVPR Sight and Sound Workshop 2024. Congratulations to Ziru!
02/2024: One paper is accepted by CVPR 2024. Congratulations to Wenjie!
Publications
-
[Under Review] Do Audio-Visual Segmentation Models Truly Segment Sounding Objects?
“AVSBench-Robust,” a comprehensive benchmark and framework addressing visual bias in audio-visual segmentation.
-
[Under Review] Robust Open-Set Test-Time Adaptation
ROSETTA decouples csID classification and csOOD detection, enhancing OOD detection and maintaining ID accuracy in open-set TTA.
-
[CVPR 2024]Segment Any Out-of-Distribution Object
”S2M”, a novel framework that transforms anomaly scores into direct segmentation prompts, eliminating the need for threshold selection, thereby improving accuracy and reducing mask fragmentation.
-
[KDD 2024 Workshop]LLM Reliable
Measuring Aleatoric and Epistemic Uncertainty in LLMs: Empirical Evaluation on ID and OOD QA Tasks.
-
[CVPR 2024 Workshop]AV-Mamba
Cross-modality selective state space models for audio-visual question answering.
Projects
-
[IAPRR Project]HAYSTAC
Systematically detects and validates trajectory anomalies through five integrated components: trip-level feature extraction, anomaly detectors, two-stage empirical calibration, nonparametric statistical scanning (NPSS), and final empirical calibration
-
[Project]EMO: Emotional Modulation and Optimization for Conversational AI
To develop EMO, we introduced a modular framework using emotional principal component vectors to manipulate hidden states, enabling precise emotional control.
