Yue Song

I am a Computing & Mathematical Sciences postdoctoral research associate at Caltech, supervised by Yisong Yue, Pietro Perona, and Max Welling. I pursued doctoral studies under European Laboratory for Learning and Intelligent Systems (ELLIS), where I was affiliated with Multimedia and Human Understanding Group (MHUG) at University of Trento, Italy and Amsterdam Machine Learning Lab (AMLab) at University of Amsterdam, the Netherlands, advised by Nicu Sebe and Max Welling.

I research structured representation learning — I am devoted to leveraging beneficial inductive biases from scientific disciplines such as math, physics, and neuroscience to improve and explain machine learning models. My current research is not task-oriented; I do not focus on a particular ML task. Instead, I am interested in developing structured methods and finding their appropriate usage in the wide application domain.

The specific deep learning fields I have worked on include high-order representation learning, decorrelated representation learning, equivariant representation learning, disentangled representation learning, and detecting/handling distribution shifts. On a theoretical aspect, the developed methodologies involve numerical and statistical matrix analysis, computational methods of matrix functions/decompositions, physics-inspired deep learning, variational inference, and matrix manifold learning.

Prior to my Ph.D. studies, I received the B.Sc. cum laude from KU Leuven, Belgium and the joint M.Sc. summa cum laude from University of Trento, Italy and KTH Royal Institute of Technology, Sweden. Besides the technical master's degree, I received an Innovation & Entrepreneurship minor degree from European Institute of Innovation and Technology (EIT Digital).

News

2024/02: I will give CVPR and ECCV tutorials on the topic of disentangled&equivariant representation learning. Stay tuned for updates!
2024/01: Our paper on Riemannian Batch Normalization has been accepted by ICLR24. Congrats Ziheng!
2023/09: Our paper on equivariant and disentangled representation learning has been accepted by NeurIPS23.
2023/05: Our paper on latent semantics discovery based on Householder transformation has been accepted by ICCV23.
2023/05: I received a gift research funding of $90,000 annually from CISCO as a co-PI.
2023/04: Our paper on latent traversal in GANs/VAEs has been accepted by ICML23.
2023/03: Our paper on jigsaw puzzle position embedding in vision Transformers has been accepted by CVPR23.
2022/12: Our paper on applying orthogonality techniques on latent disentanglement (extension of ECCV) has been accepted by IEEE T-PAMI.
2022/10: Our paper fast differentiable matrix square root and its inverse (extension of ICLR) has been accepted by IEEE T-PAMI.
2022/09: Our paper on detecting distribution shifts has been accepted by NeurIPS22.
2022/07: Two papers on the efficiency and covariance conditioning of differentiable EigenDecomposition have been accepted by ECCV22.
2022/05: Our paper on investigating the behavior of eigenvalues of global covariance pooling layer has been accepted by IEEE T-PAMI.
2022/02: Our paper on fast differentiable matrix square root has been accepted in ICLR 2022 with review score 888 (top 2.9% of submissions).
2021/07: Our paper on differentiable SVD has been accepted in ICCV21.

Selected Publications

	Unsupervised Representation Learning from Sparse Transformation Analysis Yue Song, T. Anderson Keller, Yisong Yue, Pietro Perona, Max Welling submitted to IEEE Transactions on Pattern Analysis and Machine Intelligence (T-PAMI), 2024 paper / code / bibtex
	Navigating Chemical Space with Latent Flows Guanghao Wei, Yining Huang, Chenru Duan, Yue Song♦, Yuanqi Du♦ Advances in Neural Information Processing Systems (NeurIPS), 2024 paper / code / bibtex (♦ denotes equal supervision)
	RMLR: Extending Multinomial Logistic Regression into General Geometries Ziheng Chen, Yue Song♠, Rui Wang, Xiaojun Wu, Nicu Sebe Advances in Neural Information Processing Systems (NeurIPS), 2024 paper / code / bibtex (♠ denotes the corresponding author)
	A Lie Group Approach to Riemannian Batch Normalization Ziheng Chen, Yue Song♠, Yunmei Liu, Nicu Sebe International Conference on Learning Representations (ICLR), 2024 paper / code / bibtex (♠ denotes the corresponding author)
	Flow Factorized Representation Learning Yue Song, T. Anderson Keller, Nicu Sebe, Max Welling Advances in Neural Information Processing Systems (NeurIPS), 2023 paper / code / bibtex
	Householder Projector for Unsupervised Latent Semantics Discovery Yue Song, Jichao Zhang, Nicu Sebe, Wei Wang International Conference on Computer Vision (ICCV), 2023 paper / code / bibtex
	Latent Traversals in Generative Models as Potential Flows Yue Song, T. Anderson Keller, Nicu Sebe, Max Welling International Conference on Machine Learning (ICML), 2023 paper / code / bibtex
	Masked Jigsaw Puzzle: A Versatile Position Embedding for Vision Transformers Bin Ren♣, Yahui Liu♣, Yue Song, Wei Bi, Rita Cucchiara, Nicu Sebe, Wei Wang International Conference on Computer Vision and Pattern Recognition (CVPR), 2023 paper / code / bibtex (♣ denotes co-first author)
	Orthogonal SVD Covariance Conditioning and Latent Disentanglement Yue Song, Nicu Sebe, Wei Wang IEEE Transactions on Pattern Analysis and Machine Intelligence (T-PAMI), 2022 paper / code / bibtex
	Fast Differentiable Matrix Square Root and Inverse Square Root Yue Song, Nicu Sebe, Wei Wang IEEE Transactions on Pattern Analysis and Machine Intelligence (T-PAMI), 2022 paper / code / bibtex
	RankFeat: Rank-1 Feature Removal for Out-of-distribution Detection Yue Song, Nicu Sebe, Wei Wang Advances in Neural Information Processing Systems (NeurIPS), 2022 paper / code / slides / bibtex
	Batch-efficient EigenDecomposition for Small and Medium Matrices Yue Song, Nicu Sebe, Wei Wang European Conference on Computer Vision (ECCV), 2022 paper / code / slides / bibtex
	Improving Covariance Conditioning of the SVD Meta-layer by Orthogonality Yue Song, Nicu Sebe, Wei Wang European Conference on Computer Vision (ECCV), 2022 paper / code / slides / bibtex
	On the Eigenvalues of Global Covariance Pooling for Fine-grained Visual Recognition Yue Song, Nicu Sebe, Wei Wang IEEE Transactions on Pattern Analysis and Machine Intelligence (T-PAMI), 2022 paper / code / bibtex
	Fast Differentiable Matrix Square Root Yue Song, Nicu Sebe, Wei Wang International Conference on Learning Representations (ICLR), 2022, (Top 2.9%) paper / code / slides / bibtex
	Why Approximate Matrix Square Root Outperforms Accurate SVD in Global Covariance Pooling? Yue Song, Nicu Sebe, Wei Wang International Conference on Computer Vision (ICCV), 2021 paper / supplementary / code / slides / poster / bibtex