Post-Doc Ultra-low Power Evolutionary Reinforcement Learning

This project will recruit a post-doc or engineer, depending on the candidate motivation and career, in the domain of frugal Machine Learning (ML). The aim of this position is to propose full-stack methods and open-source tools to train and infer ultra-lightweight AIs, by extending, implementing, and optimizing a new ML technique that relies on the light-by-construction and adaptive Tangled Program Graph (TPG) model.

Objectives will be adapted based on the post-doc or engineer candidate experience and interest.

The main objectives pursued are to:

• Integrate research developments of the FOUTICS project in the GEGELATI framework. Those development spans two PhD thesis on 1. Energy-aware low-complexity reinforcement learning and 2. HW-SW co-optimization for ultra-low power reinforcement learning.

• Support experiments and demonstrator development, optimize training for genetic algorithms, implement on low power microcontroler or reconfigurable devices.

• Lead new development in GEGELATI to integrate energy models and optimizations in gegelati, develop new learning environments, propose new gegelati implementation in python while keeping GEGELATI essence of fast, parallel and deterministic framework.