Description du sujet de thèse

The PhD topic is "Bioacoustic AI on a batteryless autonomous sensor".

Acoustic recording devices have a key role to play against the loss of biodiversity. Yet, the current generation of environmental sensors suffers from a lack of autonomy. Indeed, these sensors typically operate on batteries and store audio data on digital memory cards. Thus, scientists need to access the recording site periodically in order to collect data and replace batteries. Such a dependency on human intervention and on toxic substances jeopardizes the scalability of bioacoustic monitoring, disrupts natural habitats, and creates electronic waste.

In this context, PhDQ will design the first wireless sensor that embeds an AI algorithm for bioacoustic event detection. Its objectives will be:
1. Computational autonomy: the sensor will detect animal vocalizations on the sensor ("edge computing") rather than transmitting audio to a distant server ("cloud computing"). This is for reasons of energy efficiency and privacy.
2. Energetic autonomy: the sensor will be wireless and thus portable, allowing it to be deployed in remote areas.
3. Operational autonomy: the sensor will be solar-powered and batteryless, hence a reduced material footprint.

Given these objectives, the research hypothesis of the PhD is that state-of-the-art methods in AI, such as deep neural networks, must be reworked so as to satisfy the material and energetic constraints of environmental sensors. Specifically, the PhD will focus on emerging techniques in model compression, quantization, and intermittent computing.

In order to evaluate the aforementioned techniques, the PhD student will have the opportunity to conduct two field studies:
— In France, in the Atlantic Ocean, in partnership with Centrale Nantes and Open-C Foundation
— In Czech Republic, in a mountainous area, in partnership with the University of South Bohemia

A particular attention will be devoted to the measurement of electrical consumption and the life-cycle analysis of sensor parts.

Contexte de travail

The doctoral student should hold an MSc degree in computer science, mathematics, applied mathematics, statistics, artificial intelligence, or an engineering degree in a related area. Here is the expected background:
1. A profound curiosity for science is required.
2. An ability to criticize, expand, and transmit the state of the art in scientific research is required. Some experience in scientific outreach or science communication is welcome but not required.
3. Knowledge of English, both oral and written, is necessary. Knowledge of French is welcome but not necessary.
4. Basic knowledge in signal processing is necessary: e.g., short-term Fourier transform.
5. Some experience in data science, ideally speech and audio processing, is required. An experience with deep neural networks is useful but not required.
6. Basic knowledge of Python, version control (git), and command-line interfaces is required. Experience with embedded computing, high-performance computing, and parallel computing is welcome but not required.

The doctoral student will be a member of the Laboratory of Digital Sciences in Nantes (LS2N), a research unit whose components are: CNRS, Nantes University, École Centrale de Nantes, IMT Atlantique, and Inria. See: https://www.ls2n.fr/

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

The doctoral student will enroll in the "Engineering and Systems Sciences" doctoral school. See: https://ed-sis.doctorat-paysdelaloire.fr/

The doctoral student will work under the co-supervision of Vincent Lostanlen and Mathieu Lagrange, both research fellows at the CNRS. Mathieu Lagrange will be the official thesis director (in French, "habilité à diriger des recherches").

This PhD is part of the Bioacoustic AI doctoral network: visit https://bioacousticai.eu
The coordinator of the doctoral network is Dan Stowell. Consortium meetings and secondments with Pim van Gennip and Pavel Linhart are planned during the PhD.

Below is a list of potential conferences and workshops for presenting the findings of this PhD: (in alphabetical order):
- EUSIPCO
- GRETSI
- IBAC
- ICCB
- IGU
- IEEE ICASSP
- IEEE MLSP
- IEEE WASPAA

List of a potential journals : (in alphabetical order)
- EURASIP JASMP
- IEEE Sensors
- JASA
- Nature Sustainability
- Remote Sensing in Ecology and Conservation

The doctoral student will have an office shared with another student, on the site of the École Centrale de Nantes. They will have access to a work computer as well as computer equipment. They will have access to an intensive computing infrastructure, in particular GPU.

Contraintes et risques

This thesis does not involve isolated work, intense physical effort or manipulation of robotic machines.

This thesis involves a large amount of work on screen, hence some well-listed occupational risks: mainly, musculoskeletal disorders, low back pain, visual fatigue, and stress. Faced with these risks, we propose to act on the layout of the workstation, on the choice of equipment and on the organization of work. Read in particular: https://www.inrs.fr/risques/travail-ecran/ce-qu-il-faut-retenir.html (in French)

The doctoral student will not be exposed to toxic products, pathogens, noise, vibrations, dangerous radiation, or high voltage electricity.

It is expected that the doctoral student will go on a mission in France or abroad two to three times a year. The dates and destinations remain to be negotiated according to the personal constraints of the doctoral student.

The risks of falling, fire or electrocution are low, and subject to control by the laboratory's prevention assistants.

Informations complémentaires

https://bioacousticai.eu
https://cordis.europa.eu/project/id/101116715