{"type":"ScholarlyArticle","title":"Hyperbolic Knowledge Transfer with Class Hierarchy for Few-Shot Learning","authors":[{"name":"Baoquan Zhang","slug":"baoquan-zhang","given":"Baoquan","family":"Zhang","affiliation":"Harbin Institute of Technology , Shenzhen"},{"name":"Hao Jiang","slug":"hao-jiang","given":"Hao","family":"Jiang","affiliation":"Harbin Institute of Technology , Shenzhen"},{"name":"Shanshan Feng","slug":"shanshan-feng","given":"Shanshan","family":"Feng","affiliation":"Harbin Institute of Technology , Shenzhen"},{"name":"Xutao Li","slug":"xutao-li","given":"Xutao","family":"Li","affiliation":"Harbin Institute of Technology , Shenzhen"},{"name":"Yunming Ye","slug":"yunming-ye","given":"Yunming","family":"Ye","affiliation":"Harbin Institute of Technology , Shenzhen"},{"name":"Rui Ye","slug":"rui-ye","given":"Rui","family":"Ye","affiliation":"Harbin Institute of Technology , Shenzhen"}],"abstract":"Few-shot learning (FSL) aims to recognize a novel class with very few instances, which is a challenging task since it suffers from a data scarcity issue. One way to effectively alleviate this issue is introducing explicit knowledge summarized from human past experiences to achieve knowledge transfer for FSL. Based on this idea, in this paper, we introduce the explicit knowledge of class hierarchy (i.e., the hierarchy relations between classes) as FSL priors and propose a novel hyperbolic knowledge transfer framework for FSL, namely, HyperKT. Our insight is, in the hyperbolic space, the hierarchy relation between classes can be well preserved by resorting to the exponential growth characters of hyperbolic volume, so that better knowledge transfer can be achieved for FSL. Specifically, we first regard the class hierarchy as a tree-like structure. Then, 1) a hyperbolic representation learning module and a hyperbolic prototype inference module are employed to encode/infer each image and class prototype to the hyperbolic space, respectively; and 2) a novel hierarchical classification and relation reconstruction loss are carefully designed to learn the class hierarchy. Finally, the novel class prediction is performed in a nearest-prototype manner. Extensive experiments on three datasets show our method achieves superior performance over state-of-the-art methods, especially on 1-shot tasks.","doi":"10.24963/ijcai.2022/517","url":"https://ijcai.org/proceedings/2022/517","pdf":"https://ijcai.org/proceedings/2022/0517.pdf","markdown":"https://ijcai.org/proceedings/2022/517.md","bibtex":"https://ijcai.org/proceedings/2022/bibtex/517","volume":{"slug":"2022","title":"Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence","short_title":"IJCAI 2022","url":"https://ijcai.org/proceedings/2022/","year":2022,"location":"Vienna","dates":"23-29 July 2022","editors":"Luc De Raedt","isbn":"978-1-956792-00-3","publisher":"International Joint Conferences on Artificial Intelligence Organization","doi_prefix":"10.24963/ijcai.2022"},"section":"Machine Learning","track":"Main Track","number":517,"pages":{"first":3723,"last":3729},"keywords":["Machine Learning: Meta-Learning","Computer Vision: Recognition (object detection, categorization)","Machine Learning: Few-shot learning"],"videos":[{"label":"Video #1","url":"https://slideslive.com/38986799","length_sec":66,"provider":"slideslive"},{"label":"Video #2","url":"https://slideslive.com/38984942","length_sec":356,"provider":"slideslive"}],"references":[{"position":1,"raw":"[Baik et al., 2020] Sungyong Baik, Myungsub Choi, Janghoon Choi, et al. Meta-learning with adaptive hyperparameters. NeurIPS, 33:20755-20765, 2020.","title":"Meta-learning with adaptive hyperparameters","authors":"Baik","year":"2020","venue":"NeurIPS","doi":null,"url":null,"internal_url":null},{"position":2,"raw":"[Baik et al., 2021] Sungyong Baik, Janghoon Choi, Heewon Kim, et al. Meta-learning with task-adaptive loss function for few-shot learning. In ICCV, pages 9465-9474, 2021.","title":"Meta-learning with task-adaptive loss function for few-shot learning","authors":"Baik","year":"2021","venue":"ICCV","doi":null,"url":null,"internal_url":null},{"position":3,"raw":"[Chami et al., 2019] Ines Chami, Zhitao Ying, Christopher Ré, Jure Leskovec, et al. Hyperbolic graph convolutional neural networks. NeurIPS, 32:4868-4879, 2019.","title":"Hyperbolic graph convolutional neural networks","authors":"Ines Chami, Zhitao Ying, Christopher Ré, Jure Leskovec","year":"2019","venue":"NeurIPS","doi":null,"url":null,"internal_url":null},{"position":4,"raw":"[Finn et al., 2017] Chelsea Finn, Pieter Abbeel, et al. Model- agnostic meta-learning for fast adaptation of deep net- works. In Doina Precup and Yee Whye Teh, editors, ICML, volume 70, pages 1126-1135, 2017.","title":"Modelagnostic meta-learning for fast adaptation of deep networks","authors":"Finn","year":"2017","venue":null,"doi":null,"url":null,"internal_url":null},{"position":5,"raw":"[Ganea et al., 2018] Octavian-Eugen Ganea, Gary Bécigneul, Thomas Hofmann, et al. Hyperbolic neural networks. In NeurIPS, volume 31, 2018.","title":"Hyperbolic neural networks","authors":"Ganea","year":"2018","venue":"NeurIPS","doi":null,"url":null,"internal_url":null},{"position":6,"raw":"[Gao et al., 2021] Zhi Gao, Yuwei Wu, Jia, et al. Curva- ture generation in curved spaces for few-shot learning. In ICCV, pages 8691-8700, 2021.","title":"Curvature generation in curved spaces for few-shot learning","authors":"Zhi Gao, Yuwei Wu, Jia","year":"2021","venue":"ICCV","doi":null,"url":null,"internal_url":null},{"position":7,"raw":"[He et al., 2016] Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun, et al. Deep residual learning for image recognition. In CVPR, pages 770-778, 2016.","title":"Deep residual learning for image recognition","authors":"Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun","year":"2016","venue":"CVPR","doi":null,"url":null,"internal_url":null},{"position":8,"raw":"[Khrulkov et al., 2020] Valentin Khrulkov, Leyla Mir- vakhabova, Evgeniya Ustinova, Ivan Oseledets, Victor Lempitsky, et al. Hyperbolic image embeddings. In CVPR, pages 6418-6428, 2020.","title":"Hyperbolic image embeddings","authors":"Valentin Khrulkov, Leyla Mirvakhabova, Evgeniya Ustinova, Ivan Oseledets, Victor Lempitsky","year":"2020","venue":"CVPR","doi":null,"url":null,"internal_url":null},{"position":9,"raw":"[Li et al., 2020] Aoxue Li, Weiran Huang, Xu Lan, Jiashi Feng, et al. Boosting few-shot learning with adaptive mar- gin loss. In CVPR, pages 12576-12584, 2020.","title":"Boosting few-shot learning with adaptive margin loss","authors":"Li","year":"2020","venue":"CVPR","doi":null,"url":null,"internal_url":null},{"position":10,"raw":"[Liu et al., 2019a] Qi Liu, Maximilian Nickel, Douwe Kiela, et al. Hyperbolic graph neural networks. In NeurIPS, pages 8230-8241, 2019.","title":"Hyperbolic graph neural networks","authors":"Liu","year":"2019","venue":"NeurIPS","doi":null,"url":null,"internal_url":null},{"position":11,"raw":"[Liu et al., 2019b] Yanbin Liu, Juho Lee, Minseop Park, et al. Learning to propagate labels: Transductive propa- gation network for few-shot learning. In ICLR, 2019.","title":"Learning to propagate labels: Transductive propagation network for few-shot learning","authors":"Liu","year":"2019","venue":"ICLR","doi":null,"url":null,"internal_url":null},{"position":12,"raw":"[Liu et al., 2020a] Bin Liu, Yue Cao, Yutong Lin, Qi Li, Zheng Zhang, Mingsheng Long, Han Hu, et al. Negative margin matters: Understanding margin in few-shot classi- fication. In ECCV, pages 438-455, 2020.","title":"Negative margin matters: Understanding margin in few-shot classification","authors":"Liu","year":"2020","venue":"ECCV","doi":null,"url":null,"internal_url":null},{"position":13,"raw":"[Liu et al., 2020b] Shaoteng Liu, Jingjing Chen, Liangming Pan, Chong-Wah Ngo, Tat-Seng Chua, Yu-Gang Jiang, et al. Hyperbolic visual embedding learning for zero-shot recognition. In CVPR, pages 9273-9281, 2020.","title":"Hyperbolic visual embedding learning for zero-shot recognition","authors":"Liu","year":"2020","venue":"CVPR","doi":null,"url":null,"internal_url":null},{"position":14,"raw":"[Nickel et al., 2017] Maximillian Nickel, Douwe Kiela, et al. Poincaré embeddings for learning hierarchical rep- resentations. NeurIPS, 30:6338-6347, 2017.","title":"Poincaré embeddings for learning hierarchical representations","authors":"Maximillian Nickel, Douwe Kiela","year":"2017","venue":"NeurIPS","doi":null,"url":null,"internal_url":null},{"position":15,"raw":"[Peng et al., 2019] Zhimao Peng, Zechao Li, Junge Zhang, et al. Few-shot image recognition with knowledge transfer. In ICCV, pages 441-449, 2019.","title":"Few-shot image recognition with knowledge transfer","authors":"Peng","year":"2019","venue":"ICCV","doi":null,"url":null,"internal_url":null},{"position":16,"raw":"[Prabhu et al., 2019] Viraj Prabhu, Anitha Kannan, Murali Ravuri, et al. Few-shot learning for dermatological dis- ease diagnosis. In MLHC, volume 106, pages 532-552, 2019.","title":"Few-shot learning for dermatological disease diagnosis","authors":"Viraj Prabhu, Anitha Kannan, Murali Ravuri","year":"2019","venue":"MLHC","doi":null,"url":null,"internal_url":null},{"position":17,"raw":"[Qi et al., 2021] Guodong Qi, Huimin Yu, Zhaohui Lu, Shuzhao Li, et al. Transductive few-shot classification on the oblique manifold. In ICCV, pages 8412-8422, 2021.","title":"Transductive few-shot classification on the oblique manifold","authors":"Guodong Qi, Huimin Yu, Zhaohui Lu, Shuzhao Li","year":"2021","venue":"ICCV","doi":null,"url":null,"internal_url":null},{"position":18,"raw":"[Rizve et al., 2021] Mamshad Nayeem Rizve, Khan, et al. Exploring complementary strengths of invariant and equivariant representations for few-shot learning. In CVPR, pages 10836-10846, 2021.","title":"Exploring complementary strengths of invariant and equivariant representations for few-shot learning","authors":"Rizve","year":"2021","venue":"CVPR","doi":null,"url":null,"internal_url":null},{"position":19,"raw":"[Rodríguez et al., 2020] Pau Rodríguez, Issam H. Laradji, et al. Embedding propagation: Smoother manifold for few-shot classification. In ECCV, pages 121-138, 2020.","title":"Embedding propagation: Smoother manifold for few-shot classification","authors":"Pau Rodríguez, H Issam, Laradji","year":"2020","venue":"ECCV","doi":null,"url":null,"internal_url":null},{"position":20,"raw":"[Schwartz et al., 2019] Eli Schwartz, Leonid Karlinsky, et al. Baby steps towards few-shot learning with multiple se- mantics. arXiv preprint arXiv:1906.01905, 2019.","title":"Baby steps towards few-shot learning with multiple semantics","authors":"Schwartz","year":"2019","venue":null,"doi":null,"url":null,"internal_url":null},{"position":21,"raw":"[Snell et al., 2017] Jake Snell, Kevin Swersky, et al. Proto- typical networks for few-shot learning. In NeurIPS, pages 4077-4087, 2017.","title":"Prototypical networks for few-shot learning","authors":"Jake Snell, Kevin Swersky","year":"2017","venue":"NeurIPS","doi":null,"url":null,"internal_url":null},{"position":22,"raw":"[Tang et al., 2021] Shixiang Tang, Dapeng Chen, et al. Mu- tual crf-gnn for few-shot learning. In CVPR, pages 2329- 2339, June 2021.","title":"Mutual crf-gnn for few-shot learning","authors":"Tang","year":"2021","venue":"CVPR","doi":null,"url":null,"internal_url":null},{"position":23,"raw":"[Tian et al., 2020] Yonglong Tian, Yue Wang, Dilip Krish- nan, et al. Rethinking few-shot image classification: A good embedding is all you need? In ECCV, volume 12359, pages 266-282, 2020.","title":"Rethinking few-shot image classification: A good embedding is all you need?","authors":"Yonglong Tian, Yue Wang, Dilip Krishnan","year":"2020","venue":"ECCV","doi":null,"url":null,"internal_url":null},{"position":24,"raw":"[Tifrea et al., 2018] Alexandru Tifrea, Gary Bécigneul, et al. Poincar\\'e glove: Hyperbolic word embeddings. arXiv preprint arXiv:1810.06546, 2018.","title":"Poincar\\'e glove: Hyperbolic word embeddings","authors":"Alexandru Tifrea, Gary Bécigneul","year":"2018","venue":null,"doi":null,"url":null,"internal_url":null},{"position":25,"raw":"[Wan et al., 2019] Ziyu Wan, Dongdong Chen, et al. Trans- ductive zero-shot learning with visual structure constraint. In NeurIPS, pages 9972-9982, 2019.","title":"Transductive zero-shot learning with visual structure constraint","authors":"Ziyu Wan, Dongdong Chen","year":"2019","venue":"NeurIPS","doi":null,"url":null,"internal_url":null},{"position":26,"raw":"[Wertheimer et al., 2021] Davis Wertheimer, Luming Tang, et al. Few-shot classification with feature map reconstruc- tion networks. In CVPR, pages 8012-8021, 2021.","title":"Few-shot classification with feature map reconstruction networks","authors":"Davis Wertheimer, Luming Tang","year":"2021","venue":"CVPR","doi":null,"url":null,"internal_url":null},{"position":27,"raw":"[Xing et al., 2019] Chen Xing, Negar Rostamzadeh, Boris Oreshkin, Pedro O O Pinheiro, et al. Adaptive cross-modal few-shot learning. NeurIPS, 32:4847-4857, 2019.","title":"Adaptive cross-modal few-shot learning","authors":"Chen Xing, Negar Rostamzadeh, Boris Oreshkin, O O Pedro, Pinheiro","year":"2019","venue":"NeurIPS","doi":null,"url":null,"internal_url":null},{"position":28,"raw":"[Yang et al., 2020] Ling Yang, Liangliang Li, Zilun Zhang, et al. DPGN: distribution propagation graph network for few-shot learning. In CVPR, pages 13387-13396, 2020.","title":"DPGN: distribution propagation graph network for few-shot learning","authors":"Ling Yang, Liangliang Li, Zilun Zhang","year":"2020","venue":"CVPR","doi":null,"url":null,"internal_url":null},{"position":29,"raw":"[Zhang et al., 2021a] Baoquan Zhang, Ka-Cheong Leung, Xutao Li, and Yunming Ye. Learn to abstract via concept graph for weakly-supervised few-shot learning. Pattern Recognition, 117:107946, 2021.","title":"Learn to abstract via concept graph for weakly-supervised few-shot learning","authors":"Zhang","year":"2021","venue":"Pattern Recognition","doi":null,"url":null,"internal_url":null},{"position":30,"raw":"[Zhang et al., 2021b] Baoquan Zhang, Xutao Li, Yunming Ye, et al. Prototype completion with primitive knowledge for few-shot learning. In CVPR, pages 3754-3762, 2021.","title":"Prototype completion with primitive knowledge for few-shot learning","authors":"Zhang","year":"2021","venue":"CVPR","doi":null,"url":null,"internal_url":null},{"position":31,"raw":"[Zhang et al., 2022] Baoquan Zhang, Xutao Li, Yunming Ye, et al. Metanode: Prototype optimization as a neural ode for few-shot learning. In AAAI, 2022.","title":"Metanode: Prototype optimization as a neural ode for few-shot learning","authors":"Baoquan Zhang, Xutao Li, Yunming Ye","year":"2022","venue":"AAAI","doi":null,"url":null,"internal_url":null},{"position":32,"raw":"[Zheng et al., 2021] Yujia Zheng, Siyi Liu, Zekun Li, Shu Wu, et al. Cold-start sequential recommendation via meta learner. In AAAI, pages 4706-4713, 2021.","title":"Cold-start sequential recommendation via meta learner","authors":"Yujia Zheng, Siyi Liu, Zekun Li, Shu Wu","year":"2021","venue":"AAAI","doi":null,"url":null,"internal_url":null}],"license_note":"Copyright International Joint Conferences on Artificial Intelligence."}