Parameter-Efficient Interventions for Enhanced Model Merging
Fuente:
arXiv
Salvato in:
| Autori principali: | , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2024
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910759277035520 |
|---|---|
| author | Osial, Marcin Marczak, Daniel Zieliński, Bartosz |
| author_facet | Osial, Marcin Marczak, Daniel Zieliński, Bartosz |
| contents | Model merging combines knowledge from task-specific models into a unified multi-task model to avoid joint training on all task data. However, current methods face challenges due to representation bias, which can interfere with tasks performance. As a remedy, we propose IntervMerge, a novel approach to multi-task model merging that effectively mitigates representation bias across the model using taskspecific interventions. To further enhance its efficiency, we introduce mini-interventions, which modify only part of the representation, thereby reducing the additional parameters without compromising performance. Experimental results demonstrate that IntervMerge consistently outperforms the state-of-the-art approaches using fewer parameters. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2412_17023 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Parameter-Efficient Interventions for Enhanced Model Merging Osial, Marcin Marczak, Daniel Zieliński, Bartosz Computer Vision and Pattern Recognition Model merging combines knowledge from task-specific models into a unified multi-task model to avoid joint training on all task data. However, current methods face challenges due to representation bias, which can interfere with tasks performance. As a remedy, we propose IntervMerge, a novel approach to multi-task model merging that effectively mitigates representation bias across the model using taskspecific interventions. To further enhance its efficiency, we introduce mini-interventions, which modify only part of the representation, thereby reducing the additional parameters without compromising performance. Experimental results demonstrate that IntervMerge consistently outperforms the state-of-the-art approaches using fewer parameters. |
| title | Parameter-Efficient Interventions for Enhanced Model Merging |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2412.17023 |