Xinchao Wang

Deep Model Transferability from Attribution Maps

2019 · arXiv (Cornell University)

Exploring the transferability between heterogeneous tasks sheds light on their intrinsic interconnections, and consequently enables knowledge transfer from one task to another so as to reduce the training effort of the latter. In this paper, …

Structure-Aware Feature Generation for Zero-Shot Learning

2021 · arXiv (Cornell University)

Zero-Shot Learning (ZSL) targets at recognizing unseen categories by leveraging auxiliary information, such as attribute embedding. Despite the encouraging results achieved, prior ZSL approaches focus on improving the discriminant power of seen-class features, yet have …

Mosaicking to Distill: Knowledge Distillation from Out-of-Domain Data

2021 · arXiv (Cornell University)

Knowledge distillation~(KD) aims to craft a compact student model that imitates the behavior of a pre-trained teacher in a target domain. Prior KD approaches, despite their gratifying results, have largely relied on the premise that …

Reliable Label Correction is a Good Booster When Learning with Extremely Noisy Labels

2022 · arXiv (Cornell University)

Learning with noisy labels has aroused much research interest since data annotations, especially for large-scale datasets, may be inevitably imperfect. Recent approaches resort to a semi-supervised learning problem by dividing training samples into clean and …

Through the Dual-Prism: A Spectral Perspective on Graph Data Augmentation for Graph Classifications

2025 · Proceedings of the AAAI Conference on Artificial Intelligence

Graph Neural Networks (GNNs) have become the preferred tool to process graph data, with their efficacy being boosted through graph data augmentation techniques. Despite the evolution of augmentation methods, issues like graph property distortions and …

Discrete Diffusion in Large Language and Multimodal Models: A Survey

2025 · arXiv (Cornell University)

In this work, we provide a systematic survey of Discrete Diffusion Language Models (dLLMs) and Discrete Diffusion Multimodal Language Models (dMLLMs). Unlike autoregressive (AR) models, dLLMs and dMLLMs adopt a multi-token, parallel decoding paradigm using …