Tunable MAGMAX: Preference-Aware Model Merging for Continual Learning
The paper presents Tunable MAGMAX, a model merging framework designed for continual learning (CL) that accommodates user preferences. It introduces a preference vector to control task-specific performance during model merging, allowing for adaptability in various deployment environments. Experimental results indicate that Tunable MAGMAX achieves comparable or superior performance to existing methods, making it a practical solution for diverse CL applications.
- ▪Tunable MAGMAX enables preference-aware control of task-specific performance in continual learning.
- ▪The framework introduces a preference vector that adjusts the number of elements selected from each task vector during merging.
- ▪Experimental results demonstrate that Tunable MAGMAX effectively adapts merged models to various target environments.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Machine Learning arXiv:2605.20803 (cs) [Submitted on 20 May 2026] Title:Tunable MAGMAX: Preference-Aware Model Merging for Continual Learning Authors:Kei Hiroshima, Kento Uchida, Shinichi Shirakawa View a PDF of the paper titled Tunable MAGMAX: Preference-Aware Model Merging for Continual Learning, by Kei Hiroshima and 2 other authors View PDF HTML (experimental) Abstract:Continual learning (CL) aims to train models sequentially on multiple tasks while mitigating catastrophic forgetting of previously learned knowledge. Recent advances in large pre-trained models (LPMs) and model merging techniques, such as MAGMAX, have demonstrated effective CL performance by combining task-specific parameters.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
