Missions

This job opening is for postdoctoral research in fundamental data science. The primary mission is the invention of new computational methods for deep learning. In particular, work on multiresolution analysis methods, such as discrete wavelet transforms (DWT), is expected.

The primary application of this research is digital audio signal processing, with tasks such as detection, classification, clustering, segmentation, denoising, and synthesis. Extension to other time series, such as biomedical signals, may be considered.

The proposed methods must be innovative and meet specific needs in automatic audio content analysis. In particular, a recurring obstacle in the field lies in the limited amount of annotated data. It will therefore be important to implement judicious properties of inductive bias when developing deep neural network architectures.

Activities

The activities are those of a typical postdoctoral fellowship in fundamental research in computer science at the CNRS. These include: writing scientific articles, software development, performing numerical simulations, participating in team meetings, presenting work at conferences and congresses, and scientific leadership within the research community.
Teaching at the École Centrale de Nantes is encouraged but not required.

Open source library: https://github.com/kymatio/murenn

Articles already published to date: https://anr.hal.science/search/index/?q=*&anrProjectReference_s=ANR-23-CE23-0007

Skills

1. Scientific curiosity is essential.
2. An ability to critique, explore, and communicate the state of the art in research is required. Experience in scientific outreach is helpful but not required.
3. Fluency in scientific English, both written and spoken, is required. Fluency in French is helpful but not required.
4. Basic knowledge of signal processing, such as convolution, the discrete Fourier transform, and subsampling, is required. Knowledge of wavelet theory is helpful but not required.
5. Basic knowledge of probability theory, such as Gaussian vectors, the central limit theorem, and Markov's inequality, is required. Knowledge of random matrix theory and learning theory is helpful but not required.
6. Experience in data science, ideally in audio or speech processing, is required. Experience with deep neural networks is helpful but not required.
7. Ability to program in Python, use a command line, and use version control (git). Experience with embedded computing, high-performance computing (GPU-based computing), or parallel computing is helpful but not required.

Work Context

The researcher will be a member of the Nantes Digital Sciences Laboratory (LS2N), a joint research unit comprised of the CNRS, Nantes University, École Centrale de Nantes, IMT Atlantique, and Inria. See: https://www.ls2n.fr/

At LS2N, the researcher will be a member of the "Signal, Image, and Sound" (SIMS) team. See: https://audio.ls2n.fr/

The researcher will work primarily with Vincent Lostanlen and Mathieu Lagrange, both research fellows at the CNRS. Collaboration with the team's doctoral students may be considered.

This contract is part of the "Multi-Resolution Neural Networks" (MuReNN) project, funded by the French National Research Agency (ANR) for the period 2023-2027. The MuReNN project coordinator is Vincent Lostanlen. The other permanent members of the MuReNN consortium are Mathieu Lagrange, Florent de Dinechin (INSA Lyon), Anastasia Volkova (Inria Lyon), and Peter Balazs (Austrian Academy of Sciences). Research visits to Lyon and Vienna are planned during the research contract. Here is a list of relevant conferences and workshops to present the results: (in alphabetical order)
- DeepMath
- EUSIPCO
- GRETSI
- ICLR
- IEEE ICASSP
- IEEE WASPAA
- ISMIR
- MSML
- NeurIPS

Here is a list of relevant journals to present the results: (in alphabetical order)
- ACHA
- EURASIP JASMP
- IEEE SPL
- IEEE TASLP
- JASA
- JMLR
- MFML
- SIOPT

The researcher will participate in the activities of the CNRS IASIS ("Information, Learning, Signal, Image, Vision") Research Group, more specifically, its "Audio, Vision, Perception" research area. See: https://www.gdr-isis.fr/

The researcher will have an office shared with another person at the École Centrale de Nantes. He or she will have access to a work computer and hardware. He or she will have access to intensive computing infrastructure, including GPUs.

Constraints and risks

This contract does not involve isolated work, intense physical effort, or the handling of robotic machinery.
This contract involves a significant amount of screen-based work, resulting in well-documented occupational risks: primarily musculoskeletal disorders, lower back pain, visual fatigue, and stress. To address these risks, we propose to take action on workstation design, equipment selection, and work organization. See in particular: https://www.inrs.fr/risques/travail-ecran/ce-qu-il-faut-retenir.html
The researcher will not be exposed to toxic products, pathogens, noise, vibrations, hazardous radiation, or high-voltage electricity.
The researcher is expected to travel on assignment in France or abroad two to three times per year. Dates and destinations are to be negotiated based on the researcher's personal constraints. The risks of falling, fire or electrocution are low, and subject to control by the laboratory's prevention assistants.