The processing of multi-dimensional data, or tensors, has received a lot of attention from the data science community in the last decades due to the wide-ranging set of problems that are natively formulated in dimension greater than 2. For instance, the low-rank approximation problem arises in numerous applications ranging from array signal processing to data mining and data completion. Various tensor decompositions and factorization approaches have been proposed to perform denoising and reduce storage size and the complexity of further processing. Algorithmic approaches as well as analytical tools have been proposed to obtain numerical solutions to many tensor problems and characterize the properties of these solutions. Finally, the nascent field of random tensor theory builds upon random matrix theory to introduce new asymptotic results.
The Workshop on Tensor Theory and Methods, organized by Huawei Technologies France's R&D Center in Paris on November 23 and 24, 2022 gathered the top experts in the field. Presentations cover a broad scope of topics, ranging from theoretical aspects (random tensors, links with statistical physics...), algorithmic questions, to applications to wireless communications and machine learning.
While in-person participation is limited to the invited speakers, all the technical presentations were publicly live-streamed. This is our way of showing appreciation for the great work of the invited researchers, while giving back to the academic community at large. We also emphasize that all the presentations covered recent but already publicly available results.
The workshop took place at the Lagrange Mathematics and Computing Research Center, located 103 rue de Grenelle in Paris, in the historical nineteenth century building that used to terminate the Chappe optical telegraph. Note that due to space limitations, in-person attendance is very limited. If you wish to attend in person, please use the contact address at the bottom of the page.
The workshop was live-streamed (no registration required). No recordings have been made.
All speakers have kindly agreed to make their presentation material available from download. They are available by clicking the presentation titles from the schedule below.
| Time (CET) | Title | Speaker |
|---|---|---|
| 9:30 – 10:00 | Welcome Coffee (for on-site participants) | |
| 10:00 – 10:15 | Workshop Opening | Dr. Maxime Guillaud, Huawei (France) |
| 10:15 – 10:30 | Challenges of Deep Learning Algorithm and Domain Specific Silicon Architecture | Dr. Wen Tong, CTO of Huawei Wireless (Canada) |
| 10:30 – 11:20 | Tensor Modeling Based Wireless Communications | Prof. André de Almeida, Federal University of Ceará (Brazil) |
| 11:20 – 12:10 | Neural networks, flexible activation functions and tensor approximation | Dr. Konstantin Usevich, CNRS/CRAN (France) |
| 12:10 – 14:00 | Lunch (for on-site participants) | |
| 14:00 – 14:50 | Guarantees for well-posedness of canonical polyadic approximation and numerical linear algebra based estimation | Prof. Lieven De Lathauwer, KU Leuven (Belgium) |
| 14:50 – 15:40 | Tensor-Based Modulation for Massive Random Access | Dr. Maxime Guillaud, Huawei (France) |
| 15:40 – 16:00 | Coffee Break | |
| 16:00 – 16:50 | Tropical linear regression and low-rank approximation — a first step in tropical data analysis | Dr. Yang Qi, INRIA and École Polytechnique (France) |
| 16:50 – 17:40 | Nonnegative Tucker Decomposition: applications, algorithms and open questions | Dr. Jeremy Cohen, CREATIS CNRS (France) |
| 19:00 | Dinner (for on-site participants) |
| Time (CET) | Title | Speaker |
|---|---|---|
| 9:00 – 9:50 | A random matrix perspective on random tensors | Prof. Henrique de Morais Goulart, INP Toulouse (France) |
| 9:50 – 10:40 | Efficient Maximum Likelihood Estimation of a Low-Rank Probability Mass Tensor from Partial Observations | Prof. Martin Haardt, TU Ilmenau (Germany) |
| 10:40 – 11:00 | Coffee Break | |
| 11:00 – 11:50 | Coupled Tensor Decompositions for Hyperspectral Super Resolution | Prof. David Brie, CRAN, Lorraine University (France) |
| 11:50 – 12:40 | Asymptotic Analysis of Asymmetric Spiked Tensor Models with Random Matrix Theory | Dr. Mohamed El Amine Seddik, Huawei (France) |
| 12:40 | Lunch (for on-site participants) |
The workshop is organized by Huawei Technologies France's R&D Center. All enquiries shall be directed to Zhenrong LIU, zhenrong.liu@huawei.com or Maxime Guillaud, maxime.guillaud@huawei.com.