! POSITION FILLED !
Internship type
Master 2 Engineering (Master 2 or last year of an engineering school in machine learning or signal processing or biomedical engineering)
Date and duration of the internship
6 months, first semester 2025, from February to July
Context
Motor adaptation is a crucial process in human life, enabling us to adjust our movements based on sensory feedback. For instance, imagine being asked to play your favorite game of petanque with a set of balls that are heavier than usual. You would need several trials to adapt your movements and achieve the same performance. This adaptation relies on the sensory prediction error—the difference between the intended and actual outcomes on each trial. Understanding motor adaptation has broad applications, from rehabilitation in patients after brain or body injuries to optimizing motor skills in sports, work, and robotics (motor adaptation is essential for advanced dexterity).
Despite its ubiquity, motor adaptation is still not fully understood at the neural level. This project aims to bridge the gap between behaviorally-relevant variables, such as the size of errors on a given trial, and neural data obtained from magnetoencephalography (MEG) recordings.
Objective
The objective of this internship is to connect behavioral data, particularly error sizes during motor tasks, with neural data using advanced machine learning techniques. The intern will contribute to the development of a machine learning pipeline that utilizes Recurrent Neural Networks (RNNs) to analyze and predict behavior based on neural and behavioral data.
Mission
The intern will
- Develop Python code for a machine learning pipeline to process and analyze neural (magnetoencephalography data: 275 channels with a 2kHz sampling rate, recorded during previous experiments) and behavioral (joystick movements) data;
- Utilize state-of-the-art multivariate time series machine learning methods, specifically Recurrent Neural Networks (RNNs), to connect neural data with behaviorally-relevant variables, such as the size of an error on a given trial.
- Train machine learning models to explain ongoing behavior and predict future behavior based on the analyzed data.
- Gain experience in both applied machine learning and neuroscience research, enhancing their skills and CV for future academic research or ML/AI industry opportunities.
Skills required
Python, machine learning (including deep learning), pytorch, basic signal processing.
Minimal understanding of Neuroscience is desirable but not strictly necessary. Knowledge of darts/Nixtla/sktime is a plus.
Monetary compensation
Gratifications de stage (selon la réglementation en vigueur)
Contact
Dmitrii TODOROV, PhD; Chaire Professeur Junior
Neural Connectivity and Plasticity (NCP) team