At the MSAD groupe, within the LIST3N laboratory at the University of Technologi of Troyes, we are offering a research internship.

The aim of this internship is to explore and analyze a new approach based on collaborative knowledge distillation where teacher and student are trained simultaneously.

Each model mutually enriches the other through a reciprocal influence on their learning processes, going beyond traditional unidirectional transfer.

To validate this concept, the approach will be applied to speech recognition based on Connectionist Temporal Classification (CTC).

A known problem with CTC is alignment divergence: models trained separately on the same data often develop inconsistent temporal alignments.

With collaborative learning, we hypothesize that they will naturally converge towards a unified alignment, thus improving robustness and performance.

Main Tasks

• State of the Art : Analyze current research on Collaborative Knowledge Distillation (CKD), with a specific focus on its application to CTC-based speech recognition.
• Knowledge aggregation: Determine how to efficiently merge knowledge from multiple networks.
• Learning stabilization: Design regularization mechanisms to ensure a stable and balanced training process.
• Robustness assessment: Test the performance of the models under various conditions to validate their reliability.

Candidate's profile

• Education Level: Master's student (2nd year of research)
• Skills: Strong background in machine learning and deep learning.
• Development Requirements : Python proficiency and PyTorch experience.
• Preferred Assets: Interest in or experience with deep learning, Transformer architecture, speech recognition, and knowledge distillation.

Intership modalities

• Internship Start Date : September/October 2026 (depending on student’s availability)
• Internship Duration : 6 months
• Location : Université de Technologie de Troyes (UTT), France
• Language : English or French
• Expected Internship Level: M1 or M2
• Gross remuneration : 600€/month

Application

If you wish to be considered for this internship opportunity, please send your CV and cover letter to mohammed_faouzi.benzeghiba@utt.fr