Valentin Poirot

IoT and Embedded AI/ML Research Scientist - Doctoral Candidate

Kiel University, Germany & Chalmers University of Technology, Sweden

portrait

Summary

Systems Researcher with 5+ years of experience in IoT, wireless communication, and embedded artificial intelligence. I like to design TinyML solutions for small platforms using Tensorflow Lite for Microcontrollers and CMSIS-NN, and I also build distributed network protocols for IEEE 802.15.4 and Bluetooth Low Energy. I will defend my PhD in September, looking for R&D positions in embedded AI/ML starting November 2022.

Projects

On-device training and continual learning for the smallest IoT devices

How can we implement the training step of machine learning on the smallest devices with no floating-point operations, and < 100 MHz CPU and < 200 KB RAM?

BlueSeer: AI-Driven Environment Detection via BLE Scans

Environment detection and classification on embedded devices solely from Bluetooth Low Energy advertisement packets.
Published at DAC 2022, best paper candidate
[Code], [Paper]

eAFH: Improved Frequency Hopping for Bluetooth Low Energy

Adaptive channel management for Bluetooth Low Energy and Zephyr OS
Published at DCOSS 2022
[Code], [Paper]

Dimmer: RL-based Adaptive Wireless Communication

Embedded reinforcement learning and self-learned transmission policies to avoid interference in low-power IoT
Published at IEEE ICDCS 2021
[Code], [Paper], [Video]

STARC: Decentralized Coordination at Intersections

Decentralized coordination for connected cars at intersections
Published at the ITCVT Workshop at NOMS 2020
[Code], [Paper]

Wireless Paxos: Low-power Wireless Consensus

Fault-tolerant, network-wide consensus protocol for low-power IoT
Published at EWSN 2019, Best paper candidate
[Code], [Paper], [Slides]