Postdoc: Learning-based methods for large-scale imaging inverse problems (M/F)

Updated: 3 months ago
Location: Lyon 07, RHONE ALPES
Job Type: FullTime
Deadline: 19 Feb 2025

30 Jan 2025
Job Information
Organisation/Company

CNRS
Department

Laboratoire de Physique
Research Field

Engineering
Computer science
Mathematics
Researcher Profile

Recognised Researcher (R2)
Country

France
Application Deadline

19 Feb 2025 - 23:59 (UTC)
Type of Contract

Temporary
Job Status

Full-time
Hours Per Week

35
Offer Starting Date

17 Mar 2025
Is the job funded through the EU Research Framework Programme?

Not funded by a EU programme
Is the Job related to staff position within a Research Infrastructure?

No

Offer Description

Imaging inverse problems are ubiquitous across science and engineering, having a wide range of applications, from astronomical imaging to computational photography. In recent years, (deep) learning-based solutions have obtained state-of-the-art performance in many applications. However, existing methods are hard to scale to large-scale imaging problems (e.g., high-resolution images, 3D volumes, 3D + time signals) due to the large amounts of GPU memory required for training and inference [1]. Most state-of-the-art methods require that the data lies in a regular grid (pixels, voxels, etc.), which results in an exponential dependency on memory if higherresolution reconstructions are desired.
Implicit neural representations (INR) provide an alternative signal representation that can provide high-resolution reconstruction while requiring less memory to store the signal's content [2]. However, INRs have been mostly used for reconstructing a signal at a time, i.e., not leveraging any learning across a dataset of examples. Some solutions that incorporate learning have shown promising results [3,4,5]; however, it is yet unclear whether these methods can obtain a performance on par with gridbased methods and whether they can be extended to general inverse problems. The goal of this project is to study new memory (and compute) efficient learning-based reconstruction algorithms that leverage these new signal representations. Moreover, this project will adapt self-supervised learning methods [6], which do not require ground-truth data for training, to handle these new representations.

References :
1. Rudzusika, Jevgenija, et al. "3D Helical CT Reconstruction with a Memory Efficient Learned Primal-Dual Architecture." IEEE Transactions on Computational Imaging (2024).

2. Yüce, Gizem, et al. "A structured dictionary perspective on implicit neural representations." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. (2022).

3. Tancik, Matthew, et al. "Learned initializations for optimizing coordinate-based neural representations." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. (2021).

4. Chen, Yinbo, Sifei Liu, and Xiaolong Wang. "Learning continuous image representation with local implicit image function." Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2021.

5. Dupont, Emilien, et al. "From data to functa: Your data point is a function and you can treat it like one." arXiv preprint arXiv:2201.12204 (2022).

6. Tachella, Julián, Dongdong Chen, and Mike Davies. "Sensing theorems for unsupervised learning in linear inverse problems." Journal of Machine Learning Research 24.39 (2023): 1-45.

The post-doctoral researcher will work on both the theoretical understanding and development of methods based on implicit neural representations and will implement the resulting algorithms in PyTorch to test them on various inverse problems, such as image denoising or inpainting. As a secondary activity, he/she will also participate in weekly seminars and other research meetings, and will be invited to contribute to the supervision of master's student projects working on related topics.

The postdoctoral researcher will join the SYSIPHE team at the Physics Laboratory of Lyon at ENS Lyon (LPENSL), a prestigious institution focused on excellence in research. They will be part of a stimulating environment at the intersection of physics, machine learning, and signal processing, collaborating with internationally renowned researchers. ENS Lyon offers weekly seminars led by global experts, fostering rapid skill development.

What we offer:

- A stimulating work environment in close collaboration with research staff.

- 44 days of paid annual leave.

- An 18-month fixed-term contract (CDD).

- Partial reimbursement of transportation costs (75%) + a sustainable mobility allowance of up to €300/year.

- Financial contribution to health insurance costs.


Where to apply
Website
https://emploi.cnrs.fr/Candidat/Offre/UMR5672-JULTAC-003/Candidater.aspx

Requirements
Research Field
Engineering
Education Level
PhD or equivalent

Research Field
Computer science
Education Level
PhD or equivalent

Research Field
Mathematics
Education Level
PhD or equivalent

Languages
FRENCH
Level
Basic

Research Field
Engineering
Years of Research Experience
1 - 4

Research Field
Computer science
Years of Research Experience
1 - 4

Research Field
Mathematics
Years of Research Experience
1 - 4

Additional Information
Eligibility criteria

- A PhD thesis on imaging inverse problems, computer vision or signal processing.

- Basic knowledge of deep learning libraries such as PyTorch or JAX.

- Publications in top imaging journals (e.g., IEEE TCI, TIP, TSP, TPAMI, SIAM Imag. Sciences) or in top CV/ML conferences (e.g., CVPR, ICCV, NeurIPS, ICML, ICLR, AISTATS).


Website for additional job details

https://emploi.cnrs.fr/Offres/CDD/UMR5672-JULTAC-003/Default.aspx

Work Location(s)
Number of offers available
1
Company/Institute
Laboratoire de Physique
Country
France
City
LYON 07
Geofield


Contact
City

LYON 07
Website

http://www.ens-lyon.fr/PHYSIQUE/

STATUS: EXPIRED

  • X (formerly Twitter)
  • Facebook
  • LinkedIn
  • Whatsapp

  • More share options
    • E-mail
    • Pocket
    • Viadeo
    • Gmail
    • Weibo
    • Blogger
    • Qzone
    • YahooMail