ANDONG WANG'S HOMEPAGE
Research Scientist · RIKEN AIP

Andong Wang 王安东

Andong is a researcher at RIKEN AIP studying the so-called tensor learning, often by extending ideas from matrices and functions to tensors. He hopes to develop ideas that are both conceptually deep and practically useful.

Selected Publications
Tensor Learning Algebraic & Geometric Methods Trustworthy Learning Quantum Machine Learning Negative Obstacle Detection for UGV Intelligent Welding

Research Directions

A compact view of research themes around representation, robustness, transfer, and reliable learning from imperfect information.

Data spaces · Tensor geometry

Algebraic and geometric structures induced by the t-product and transform-domain tensor operations, toward a geometric view of tensor learning through t-scalars, t-modules, and t-manifolds.

Parameter spaces · Low-rank robustness

Transformed low-rank parameterization for robust generalization, adversarial robustness, sample efficiency, and transferability in tensor neural networks and tensor regression models.

Function spaces · Functional tensor SVD

Extensions of tensor decomposition/learning from discrete arrays to function spaces for functional data, multi-output learning, operator learning, lifelong learning, etc., with a focus on efficient and robust generalization under distribution shift.

Applications · Quantum ML and Multi-Modal ML

Quantum machine learning and Multi-Modal learning, where tensor networks provide a compact language for structured representations, prompt search, adversarial robustness, and learning from noisy or incomplete information.

Education and Research Experience

Education and research appointments.

Research support: JSPS Early-Career Grant, KAKENHI 25K21283, 2025–2028. Selected recognition: RIKEN Research and Technology Incentive Award (Ohbu, 理研樱舞奖), 2025.

Academic Service

Selected organization, service, and reviewing activities.

  1. Secretary, RIKEN Researcher Assembly（理化学研究所研究員会議幹事）, FY2026–FY2027.
  2. Area Chair, International Conference on Machine Learning (ICML), 2026.
  3. Co-organizer, ICIAM 2023 Mini-symposium: New Trends in Tensor Networks and Tensor Optimization.
  4. Workshop Co-Chair, IEEE CAI 2024 Workshop: Tensor Models for Machine Learning.
  5. Reviewer for NeurIPS, ICML, ICLR, AISTATS, AAAI, IJCAI, CVPR, TPAMI, TIP, TSP, TNNLS, TCYB, TCSVT, SIIMS, Pattern Recognition, Neural Networks, and Machine Learning.

Selected Publications

Selected papers on tensor geometry, robust generalization, transfer learning, and tensor recovery. 拿着张量这个锤子，他看啥都像钉子，到处锤阿锤啊锤，于是自己也成了个锤子。如果论文有误，希望邮件教育（w.a.d@outlook.com）。

  1. Refining dual spectral sparsity in transformed tensor singular values.
     Wang, A., Qiu, Y., Huang, H., Jin, Z., Zhou, G., & Zhao, Q. · ICML 2026. Previous OpenReview Submission
  2. Towards a geometric understanding of tensor learning via the t-product.
     Wang, A., Qiu, Y., Huang, H., Jin, Z., Zhou, G., & Zhao, Q. · NeurIPS 2025. PDF Code
  3. Low-Rank Tensor Transitions (LoRT) for transferable tensor regression.
     Wang, A., Qiu, Y., Jin, Z., Zhou, G., & Zhao, Q. · ICML 2025. PDF Code
  4. Generalized tensor decomposition for understanding multi-output regression under combinatorial shifts.
     Wang, A., Qiu, Y., Bai, M., Jin, Z., Zhou, G., & Zhao, Q. · NeurIPS 2024. PDF Code
  5. Transformed low-rank parameterization can help robust generalization for tensor neural networks.
     Wang, A., Li, C., Bai, M., Jin, Z., Zhou, G., & Zhao, Q. · NeurIPS 2023. PDF Code
  6. Tensor recovery via L-spectral k-support norm.
     Wang, A., Zhou, G., Jin, Z., & Zhao, Q. · IEEE Journal of Selected Topics in Signal Processing, 15(3), 522–534, 2021. Link
  7. Robust Tensor Decomposition via Orientation Invariant Tubal Nuclear Norms.
     Wang, A., Li, C., Jin, Z., & Zhao, Q. · AAAI 2020, Oral. PDF Code
  8. Latent Schatten TT norm for tensor completion.
     Wang, A., Song, X., Wu, X., Lai, Z., & Jin, Z. · ICASSP 2019, Oral. Link
  9. Noisy low-tubal-rank tensor completion.
     Wang, A., Lai, Z., & Jin, Z. · Neurocomputing, 330, 267–279, 2019. Link
  10. Near-optimal noisy low-tubal-rank tensor completion via singular tube thresholding.
     Wang, A. & Jin, Z. · ICDM Workshop 2017. Link

Reflections from More Than a Decade of Research

Some reflections after years of research. I am not a successful person, and perhaps not qualified to write these words. I only hope to encourage those whose work is still unseen. If your work is hard, slow, and not easily rewarded, it does not mean it has no value. Perhaps you are doing something meaningful before the world has learned how to see it. I also began on a less visible path, and after five years of being worn down by reality, slowly learned to choose what could be seen first. That choice was practical, but it made me more indebted to what remains unseen.

What Becomes Visible

This may be a kind of survivorship bias. More tractable, faster, and more publishable problems often rise to the center of the visible system, partly because they fit short-term metrics, stable benchmarks, and results that can be neatly packaged, and partly because the system often rewards what is easiest to count rather than what is hardest to value. In many environments, such work is quickly judged as success: a paper is accepted, a grant is funded, a tenure-track review is passed. Gradually, it becomes easier to keep doing this research.

What Remains Unseen

This may also be a kind of survivorship bias. Less tractable, deeper, and perhaps more ambitious problems often stay outside the visible system, not because they lack value, but because they lack short-term metrics, stable benchmarks, or results that can be neatly packaged. In some environments, such work is easily judged as failure: a PhD may not be finished, a grant may not be funded, a tenure-track review may not be passed. Slowly, it becomes much harder to continue doing research.

A Difficult Choice. What becomes visible as success in research is not always identical to what will remain valuable in the long run. It is often shaped by timing, presentation, community attention, and the publication system through which ideas must pass.
Life is short, and we all carry many responsibilities, both in research and in life, some of which inevitably conflict with one another. Choosing what to pursue in research therefore remains a deeply difficult question.

十年间，想搞定负障碍，一直不得其法。

几年来，想学代数几何，却是一塌糊涂。

近两年，想建模人与AI，难逃思维桎梏。

。。。。。。

东子人菜瘾大，谨小慎微，也经常出错。

曾是太平天国迷，常回想平在山兄弟们二十岁的旷世呐喊，而自己年近四十、须发渐白，却还在逃避现实困难，靠涂几页薄纸取巧偷安。

。。。。。。

稀里糊涂二十年，对人对己对科研，想轻轻说一句：

进步太慢很抱歉

仍似少年磨长剑