1 code implementation • ICML 2020 • chengyu dong, Liyuan Liu, Zichao Li, Jingbo Shang
Serving as a crucial factor, the depth of residual networks balances model capacity, performance, and training efficiency.
no code implementations • EMNLP 2021 • Zihan Wang, chengyu dong, Jingbo Shang
In this paper, we present an empirical property of these representations—”average” approximates “first principal component”.
no code implementations • 12 Nov 2023 • Xiyuan Zhang, Xiaohan Fu, Diyan Teng, chengyu dong, Keerthivasan Vijayakumar, Jiayun Zhang, Ranak Roy Chowdhury, Junsheng Han, Dezhi Hong, Rashmi Kulkarni, Jingbo Shang, Rajesh Gupta
By obviating the need for ground truth clean data, our method offers a practical denoising solution for real-world applications.
no code implementations • 11 Oct 2023 • chengyu dong, Liyuan Liu, Hao Cheng, Jingbo Shang, Jianfeng Gao, Xiaodong Liu
Although ELECTRA offers a significant boost in efficiency, its potential is constrained by the training cost brought by the auxiliary model.
no code implementations • 4 Oct 2023 • An Yan, Yu Wang, Yiwu Zhong, Zexue He, Petros Karypis, Zihan Wang, chengyu dong, Amilcare Gentili, Chun-Nan Hsu, Jingbo Shang, Julian McAuley
Medical image classification is a critical problem for healthcare, with the potential to alleviate the workload of doctors and facilitate diagnoses of patients.
1 code implementation • ICCV 2023 • An Yan, Yu Wang, Yiwu Zhong, chengyu dong, Zexue He, Yujie Lu, William Wang, Jingbo Shang, Julian McAuley
Recent advances in foundation models present new opportunities for interpretable visual recognition -- one can first query Large Language Models (LLMs) to obtain a set of attributes that describe each class, then apply vision-language models to classify images via these attributes.
1 code implementation • 24 May 2023 • Dheeraj Mekala, Adithya Samavedhi, chengyu dong, Jingbo Shang
To address the annotation bottleneck, we introduce SELFOOD, a self-supervised OOD detection method that requires only in-distribution samples as supervision.
1 code implementation • 24 May 2023 • chengyu dong, Zihan Wang, Jingbo Shang
We show that the limited performance of seed matching is largely due to the label bias injected by the simple seed-match rule, which prevents the classifier from learning reliable confidence for selecting high-quality pseudo-labels.
no code implementations • 2 Feb 2023 • chengyu dong
Deep neural networks have seen enormous success in various real-world applications.
no code implementations • 14 Jun 2022 • chengyu dong, Liyuan Liu, Jingbo Shang
How to conduct teacher training for knowledge distillation is still an open problem.
1 code implementation • 25 May 2022 • Dheeraj Mekala, chengyu dong, Jingbo Shang
Weakly supervised text classification methods typically train a deep neural classifier based on pseudo-labels.
no code implementations • 16 Dec 2021 • Mingqi Lv, chengyu dong, Tieming Chen, Tiantian Zhu, Qijie Song, Yuan Fan
To effective and efficient detect cyber-attacks from a huge number of system events in the provenance data, we firstly model the provenance data by a heterogeneous graph to capture the rich context information of each system entities (e. g., process, file, socket, etc.
no code implementations • 7 Oct 2021 • chengyu dong, Liyuan Liu, Jingbo Shang
We show that label noise exists in adversarial training.
no code implementations • 29 Sep 2021 • Zichao Li, Liyuan Liu, chengyu dong, Jingbo Shang
While this phenomenon is commonly explained as overfitting, we observe that it is a twin process: not only does the model catastrophic overfits to one type of perturbation, but also the perturbation deteriorates into random noise.
no code implementations • Findings (EMNLP) 2021 • Zichao Li, Dheeraj Mekala, chengyu dong, Jingbo Shang
To recognize the poisoned subset, we examine the training samples with these identified triggers as the most suspicious token, and check if removing the trigger will change the poisoned model's prediction.
1 code implementation • 18 Apr 2021 • Zihan Wang, chengyu dong, Jingbo Shang
In this paper, we present an empirical property of these representations -- "average" approximates "first principal component".
1 code implementation • 15 Feb 2021 • chengyu dong, Liyuan Liu, Jingbo Shang
Specifically, we first propose a strategy to measure the data quality based on the learning behaviors of the data during adversarial training and find that low-quality data may not be useful and even detrimental to the adversarial robustness.
2 code implementations • 15 Oct 2020 • Zichao Li, Liyuan Liu, chengyu dong, Jingbo Shang
Our goal is to understand why the robustness drops after conducting adversarial training for too long.