Moteur de recherche d'offres d'emploi CEA

POST DOC Spiking Neural Network incremental learning algorithm for neuroprosthetics application

Vacancy details

General information


The French Alternative Energies and Atomic Energy Commission (CEA) is a key player in research, development and innovation in four main areas :
• defence and security,
• nuclear energy (fission and fusion),
• technological research for industry,
• fundamental research in the physical sciences and life sciences.

Drawing on its widely acknowledged expertise, and thanks to its 16000 technicians, engineers, researchers and staff, the CEA actively participates in collaborative projects with a large number of academic and industrial partners.

The CEA is established in ten centers spread throughout France



Position description


Engineering science



Job title

POST DOC Spiking Neural Network incremental learning algorithm for neuroprosthetics application


To reach the goal of restoring the autonomy of spinal cord injury patients at home, a fully implantable neuroprosthetics solution must be developed. A major milestone was recently reached by CEA with the first successful demonstration of a functional neuroprosthetics: a tetraplegic patient was able to control an exoskeleton with 8D control, based on electrocorticography (ECoG) with a wireless recording device. This achievement was made in a laboratory environment. The next key milestone is to move out of the lab to ensure patient long-term autonomy.
An ultra-low power Neuronal Processing Unit (NPU), ideally embedded in the prosthetics, is critically needed. The NeMo project will focus on (1) developing adaptive decoding algorithms for neuroprosthetics, (2) taking into account the constraints of ultra-low power, embedded, NPU design (e.g. memory footprint, weight quantization) but also robustness to faults.

Contract duration (months)


Job description

The key scientific objective of NeMo project is to develop online, adapti[TA21] ve, low-power decoding and regression algorithms. Since those algorithms will be executed on a Neural Processing Unit (outside the scope of this project), great care will be taken to ensure that (1) they are compatible with a hardware implementation and (2) they can be executed in a real-time fashion (decision rate @ 10Hz and update rate @ 0.5Hz-1Hz).

The Post-doctoral research work is planned over 2 years and can be broken down into the following tasks:


Task 1: SOA analysis – 4 months

The first 4 months will be dedicated to state-of-the-art review and update (part of this SOA is already well known to the consortium), especially looking for the latest developments in knowledge representation in SNNs for Brain-Computer Interfaces, online learning and evolving SNN topologies (population coding) for dealing with drifting data streams.


Task 2: Initial learning strategy experimentation – 4 months

In this task, the consortium will select the most promising algorithms, from the SOA analysis and will experiment with them offline using the data base. The objective is to get clues, very early on, on whether those algorithms will scale well on problem with a higher number of Degrees of Freedom. The best one will be kept for the next Task.


Task 3: Algorithm development – 8 months

This will be the core of the project, refining the most appropriate algorithm to fit the needs of BCI processing. Special care will be given to real time operating capabilities, taking into account the constraints of hardware acceleration (memory capacity, quantization …).


Task 4: Implementation, integration to experimental environment and test – 6 months

This task will be pursued only if the previous ones prove to be highly successful. The developed algorithm will be experimented live. To do this, the developed algorithms will be implemented so as to be compatible with real-time use and integrated to the ABSD (Adaptive Brain Signal Decoder) software environment and experimental chain of “BCI and Tetraplegia” Clinical Trial. Finally, it will be tested in BCI experiments.


Task 5: System specification – 2 months

This project will end with a specification of the system that will be capable of implementing the algorithm, especially the Neural Processing Unit: number of parameters to store, number of operations to perform per second, types of operators to implement, maximum power dissipation …


The post-doctoral student will have access to world-class research facilities and will be integrated in a highy skilled and interdisciplinary team.



Applicant Profile

Technical skills:

Sound mathematical knowledge (applied mathematics, signal processing, eventually knowledge of Matlab), very good algorithmic and programming levels (Python, C/C++, scripting), goodknowledge of machine learning algorithms and frameworks. A knowledge of data science or cognitive science would be a plus


Personal skills:

Determination, perseverance, trustworthiness, autonomy, adaptability, initiative, good communication skills



English: at least B2 equivalent, excellent reading and writing level, good speaking level

French: Fluency in French is a plus, but it is not mandatory.


Taking up office before February / March 2022 is imperative.

Position location





Candidate criteria


English (Fluent)