小样本机器学习
简介:
以大模型为代表的深度学习技术已取得巨大突破,但在具体应用时往往面临困难,具体表现为:(1)与通用场景跨度大的专用场景下性能差,例如医疗、遥感等场景;(2)通用场景下数据本就稀少的任务中性能差,例如异常行为等。为解决此问题,现有通用做法为进一步增大数据集,为每个专用领域训练独有的大模型;但由于专用领域层出不穷,这无疑是费时费力的。与此相比,人类天然拥有良好的小样本学习能力,无需海量数据即可在每个领域下实现良好的学习效果。人类这种能力表现为对已学习知识进行可解释的迁移、重组、拓展。为模仿人类这种学习能力,小样本学习的研究集中于多模态大模型的迁移性、重组性、拓展性、可解释性等方面的研究。小样本学习组的负责老师为邹逸雄老师(主页:/hust/smu_idc/info/1021/1843.htm)。
代表性论文:
1. Yixiong Zou, Shuai Yi, Yuhua Li, Ruixuan Li; A Closer Look at the CLS Token for Cross-Domain Few-Shot Learning, Annual Conference on Neural Information Processing Systems (NeurIPS 2024, CCF-A类论文), Vancouver, CA, 2024-12-15 - 2024-12-15
2. Yixiong Zou, Ran Ma, Yuhua Li, Ruixuan Li; Attention Temperature Matters in ViT-Based Cross-Domain Few-Shot Learning, Annual Conference on Neural Information Processing Systems (NeurIPS 2024, CCF-A类论文), Vancouver, CA, 2024-12-15 - 2024-12-15
3. Yiman Hu, Yixiong Zou (通讯), Ruixuan Li, Yuhua Li; Generate Universal Adversarial Perturbations for Few-Shot Learning, Annual Conference on Neural Information Processing Systems (NeurIPS 2024, CCF-A类论文), Vancouver, CA, 2024-12-15 - 2024-12-15
4. Jintao Tong, Yixiong Zou (通讯), Yuhua Li, Ruixuan Li; Lightweight Frequency Masker for Cross-Domain Few-Shot Semantic Segmentation, Annual Conference on Neural Information Processing Systems (NeurIPS 2024, CCF-A类论文), Vancouver, CA, 2024-12-15 - 2024-12-15
5. Yixiong Zou, Shanghang Zhang, Haichen Zhou, Yuhua Li, Ruixuan Li; Compositional Few-Shot Class-Incremental Learning. International Conference on Machine Learning (ICML 2024, CCF-A类论文), Vienna, Austria, July 21-27, 2024
6. Yixiong Zou, Yicong Liu, Yiman Hu, Yuhua Li, Ruixuan Li; Flatten Long-Range Loss Landscapes for Cross-Domain Few-Shot Learning. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2024, CCF-A类论文), Seattle, USA, June 17-21, 2024
7. Haichen Zhou, Yixiong Zou (通讯), Ruixuan Li, Yuhua Li, Kui Xiao; Delve into Base-Novel Confusion: Redundancy Exploration for Few-Shot Class-Incremental Learning. International Joint Conference on Artificial Intelligence (IJCAI 2024, CCF-A类论文), Jeju Island, South Korea, August 3-9, 2024.
8. Jie Han, Yixiong Zou (通讯), Haozhao Wang, Wei Liu, Yao Wu, Tao Zhang, Ruixuan Li; Decoupling Representation and Knowledge for Few-Shot Intent Classification and Slot Filling. Annual AAAI Conference on Artificial Intelligence (AAAI 2024, CCF-A类论文), Vancouver, Canada, February 20th-27th, 2024.
9. Zhenyu Zhang, Guangyao Chen, Yixiong Zou (通讯), Yuhua Li, Ruixuan Li; Learning Unknowns from Unknowns: Diversified Negative Prototypes Generator for Few-shot Open-Set Recognition. ACM International Conference on Multimedia (ACM MM 2024, CCF-A类论文), Melbourne, Australia, Oct 28 - Nov 1, 2024.
10. Zhenyu Zhang, Guangyao Chen, Yixiong Zou (通讯), Zhimeng Huang, Yuhua Li, Ruixuan Li; MICM: Rethinking Unsupervised Pretraining for Enhanced Few-shot Learning. ACM International Conference on Multimedia (ACM MM 2024, Oral, CCF-A类论文), Melbourne, Australia, Oct 28 - Nov 1, 2024.
11. Yixiong Zou, Shanghang Zhang, Yuhua Li, Ruixuan Li; Margin-Based Few-Shot Class-Incremental Learning with Class-Level Overfitting Mitigation, Annual Conference on Neural Information Processing Systems (NeurIPS 2022, CCF-A类论文), New Orleans, LA, USA, 2022-11-28 - 2022-12-3
12. Yixiong Zou, Shanghang Zhang, Jianpeng Yu, Yonghong Tian, José M. F. Moura; Revisiting Mid-Level Patterns for Cross-Domain Few-Shot Recognition, ACM International Conference on Multimedia (ACM MM 2021, CCF-A类论文), Chengdu, China, 2021-10-20 - 2021-10-24
13. Yixiong Zou, Shanghang Zhang, Guangyao Chen, Yonghong Tian, Kurt Keutzer, José M. F. Moura; Annotation-Efficient Untrimmed Video Action Recognition, ACM International Conference on Multimedia (ACM MM 2021, CCF-A类论文), Chengdu, China, 2021-10-20 - 2021-10-24
14. Yixiong Zou, Shanghang Zhang, Ke Chen, Yonghong Tian, Yaowei Wang, José M. F. Moura; Compositional Few-Shot Recognition with Primitive Discovery and Enhancing, ACM International Conference on Multimedia (ACM MM 2020, Oral, CCF-A类论文), Seattle, WA, USA, 2020-10-12 - 2020-10-16
15. Yicong Liu, Yixiong Zou (通讯), Ruixuan Li, Yuhua Li; Spectral Decomposition and Transformation for Cross-Domain Few-Shot Learning, Neural Networks (CCF-B类论文, 一区SCI期刊), 2024, https://doi.org/10.1016/j. neunet.2024.106536
16. Yixiong Zou, Yemin Shi, Daochen Shi, Yaowei Wang, Yongsheng Liang, Yonghong Tian; Adaptation-Oriented Feature Projection for One-shot Action Recognition, IEEE Transactions on Multimedia (TMM 2020, CCF-B类论文, 一区SCI期刊), 2020, 22(12): 3166-3179
其他论文:
请见 https://scholar.google.com/citations?user=VXxF0mcAAAAJ&hl=zh-CN
科研项目:
面向开放世界的模块化终生学习方法研究(国家自然科学基金)(2023-2025)
竞赛获奖:
2023年第五届全球校园人工智能算法精英大赛 - AIGC驱动的小样本分类 - 全国一等奖