Zihao Zhu is currently a PhD student in Data Science at The Chinese University of Hong Kong, Shenzhen (CUHKSZ). His primary research interests are Trustworthy AI, Backdoor Attack and Defense in Deep Learning. He has published related papers in top tier conferences and journals such as NeurIPS, IJCAI, ICASSP and PR.
Ph.D. in Data Science, 2021-
The Chinese University of Hong Kong, Shenzhen
MSc in CS, 2018-2021
University of Chinese Academy of Sciences
BSc in CS, 2014-2018
China University of Mining and Technology
The paper highlights the importance of detecting dirty samples in datasets used for Data-centric AI, such as poisoned samples or noisy labels, which compromise the reliability of DNNs. To address this, we propose the Versatile Data Cleanser (VDC), a novel approach leveraging multimodal large language models to identify inconsistencies between images and their labels, demonstrating its effectiveness across various types of dirty samples.
The paper introduces a multi-modal heterogeneous graph-based approach with a modality-aware graph convolutional network for Fact-based Visual Question Answering, enhancing the selection and aggregation of relevant evidence across modalities, thereby achieving state-of-the-art performance and interpretability.