Hendijani Fard, Fatemeh et Ouni, Ali.
2026.
« Scaling code assistants without retraining with model merging: Capability fusion & sustainable adaptation ».
In FSE Companion 2026: Proceedings of the 34th ACM International Conference on the Foundations of Software Engineering (Montreal, QC, Canada, July 05-09, 2026)
62–63.
Association for Computing Machinery, Inc.
Prévisualisation |
PDF
Ouni-A-2026-34230.pdf - Version publiée Licence d'utilisation : Creative Commons CC BY. Télécharger (390kB) | Prévisualisation |
Résumé
Model merging, is a techniques that integrates AI-based models such as language or vision models using a variety of techniques, with or without additional training. Model merging has been used for multi-task learning, unlearning, and knowledge transfer. Several merged models on Open LLM leaderboards proves its success in various domains. However, less attention is paid to this technique in software engineering. In this technical briefing, we will discuss the main concepts and approaches for model merging, and then explore rare related studies in software engineering, including our experiments of using merged models for automatic program repair and generalizability of the merged model on this task. We will conclude by future directions and research opportunities of merged models in software engineering research. This briefing can attract audience from academia and industry and no specific background is needed.
| Type de document: | Compte rendu de conférence |
|---|---|
| Chercheur(-euse): | Chercheur(-euse) Ouni, Ali |
| Affiliation: | Génie logiciel et des technologies de l'information |
| Date de dépôt: | 12 août 2026 19:04 |
| Dernière modification: | 26 sept. 2026 18:53 |
| URI: | https://espace2.etsmtl.ca/id/eprint/34230 |
Actions (Authentification requise)
 |
Dernière vérification avant le dépôt |