Timezone: »
Deep learning models have been found with a tendency of relying on shortcuts, i.e., decision rules that perform well on standard benchmarks but fail when transferred to more challenging testing conditions. Such reliance may hinder deep learning models from learning other task-related features and seriously affect their performance and robustness. Although recent studies have shown some characteristics of shortcuts, there are few investigations on how to help the deep learning models to solve shortcut problems. This paper proposes a framework to address this issue by setting up roadblocks on shortcuts. Specifically, roadblocks are placed when the model is urged to learn to complete a gently modified task to ensure that the learned knowledge, including shortcuts, is insufficient the complete the task. Therefore, the model trained on the modified task will no longer over-rely on shortcuts. Extensive experiments demonstrate that the proposed framework significantly improves the training of networks on both synthetic and real-world datasets in terms of both classification accuracy and feature diversity. Moreover, the visualization results show that the mechanism behind the proposed our method is consistent with our expectations. In summary, our approach can effectively disable the shortcuts and thus learn more robust features.
Author Information
Hongjing Niu (University of Science and Technology of China)
Hanting Li (University of Science and Technology of China)
Feng Zhao (University of Science and Technology of China)
Bin Li (University of Science and Technology of China)
More from the Same Authors
-
2023 Poster: FouriDown: Factoring Down-Sampling into Shuffling and Superposing »
Qi Zhu · man zhou · Jie Huang · Naishan Zheng · Hongzhi Gao · Chongyi Li · Yuan Xu · Feng Zhao -
2023 Poster: Domain Re-Modulation for Few-Shot Generative Domain Adaptation »
Yi Wu · Ziqiang Li · Chaoyue Wang · Heliang Zheng · Shanshan Zhao · Bin Li · Dacheng Tao -
2023 Poster: Deep Fractional Fourier Transform »
Hu Yu · Jie Huang · Lingzhi LI · man zhou · Feng Zhao -
2023 Poster: Transition-constant Normalization for Image Enhancement »
Jie Huang · man zhou · Jinghao Zhang · Gang Yang · Mingde Yao · Chongyi Li · Zhiwei Xiong · Feng Zhao -
2022 Spotlight: Lightning Talks 2B-4 »
Feiyi Xiao · Amrutha Saseendran · Kwangho Kim · Keyu Yan · Changjian Shui · Guangxi Li · Shikun Li · Edward Kennedy · Man Zhou · Gezheng Xu · Ruilin Ye · Xiaobo Xia · Junjie Tang · Kathrin Skubch · Stefan Falkner · Hansong Zhang · Jose Zubizarreta · Huaying Fang · Xuanqiang Zhao · Jie Huang · Qi CHEN · Yibing Zhan · Jiaqi Li · Xin Wang · Ruibin Xi · Feng Zhao · Margret Keuper · Charles Ling · Shiming Ge · Chengjun Xie · Tongliang Liu · Tal Arbel · Chongyi Li · Danfeng Hong · Boyu Wang · Christian Gagné -
2022 Spotlight: Deep Fourier Up-Sampling »
man zhou · Hu Yu · Jie Huang · Feng Zhao · Jinwei Gu · Chen Change Loy · Deyu Meng · Chongyi Li -
2022 Spotlight: Panchromatic and Multispectral Image Fusion via Alternating Reverse Filtering Network »
Keyu Yan · Man Zhou · Jie Huang · Feng Zhao · Chengjun Xie · Chongyi Li · Danfeng Hong -
2022 Spotlight: Lightning Talks 2B-1 »
Yehui Tang · Jian Wang · Zheng Chen · man zhou · Peng Gao · Chenyang Si · SHANGKUN SUN · Yixing Xu · Weihao Yu · Xinghao Chen · Kai Han · Hu Yu · Yulun Zhang · Chenhui Gou · Teli Ma · Yuanqi Chen · Yunhe Wang · Hongsheng Li · Jinjin Gu · Jianyuan Guo · Qiman Wu · Pan Zhou · Yu Zhu · Jie Huang · Chang Xu · Yichen Zhou · Haocheng Feng · Guodong Guo · yongbing zhang · Ziyi Lin · Feng Zhao · Ge Li · Junyu Han · Jinwei Gu · Jifeng Dai · Chao Xu · Xinchao Wang · Linghe Kong · Shuicheng Yan · Yu Qiao · Chen Change Loy · Xin Yuan · Errui Ding · Yunhe Wang · Deyu Meng · Jingdong Wang · Chongyi Li -
2022 Poster: Deep Fourier Up-Sampling »
man zhou · Hu Yu · Jie Huang · Feng Zhao · Jinwei Gu · Chen Change Loy · Deyu Meng · Chongyi Li -
2022 Poster: Panchromatic and Multispectral Image Fusion via Alternating Reverse Filtering Network »
Keyu Yan · Man Zhou · Jie Huang · Feng Zhao · Chengjun Xie · Chongyi Li · Danfeng Hong