PiKE: Adaptive Data Mixing for Large-Scale Multi-Task Learning Under Low Gradient Conflicts
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2025
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866916776783118336 |
|---|---|
| author | Li, Zeman Deng, Yuan Zhong, Peilin Razaviyayn, Meisam Mirrokni, Vahab |
| author_facet | Li, Zeman Deng, Yuan Zhong, Peilin Razaviyayn, Meisam Mirrokni, Vahab |
| contents | Modern foundation models are trained on diverse datasets to enhance generalization across tasks and domains A central challenge in this process is determining how to effectively mix and sample data from multiple sources This naturally leads to a multitask learning (MTL) perspective While prior work in MTL has emphasized mitigating gradient conflicts we observe that largescale pretraining scenariossuch as multilingual or multidomain trainingoften exhibit little to no gradient conflict Motivated by this observation we propose PiKE (Positive gradient interaction-based K-task weights Estimator) an adaptive data mixing algorithm that dynamically adjusts sampling weights during training PiKE exploits nonconflicting gradient interactions to minimize a neartight upper bound on the average loss decrease at each step while incurring negligible computational overhead We provide theoretical convergence guarantees and show that PiKE outperforms static and nonadaptive mixing baselines Furthermore we extend PiKE to promote balanced learning across tasks Extensive experiments on largescale language model pretraining confirm that PiKE achieves faster convergence and improved downstream performance compared to existing approaches |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_06244 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | PiKE: Adaptive Data Mixing for Large-Scale Multi-Task Learning Under Low Gradient Conflicts Li, Zeman Deng, Yuan Zhong, Peilin Razaviyayn, Meisam Mirrokni, Vahab Machine Learning Modern foundation models are trained on diverse datasets to enhance generalization across tasks and domains A central challenge in this process is determining how to effectively mix and sample data from multiple sources This naturally leads to a multitask learning (MTL) perspective While prior work in MTL has emphasized mitigating gradient conflicts we observe that largescale pretraining scenariossuch as multilingual or multidomain trainingoften exhibit little to no gradient conflict Motivated by this observation we propose PiKE (Positive gradient interaction-based K-task weights Estimator) an adaptive data mixing algorithm that dynamically adjusts sampling weights during training PiKE exploits nonconflicting gradient interactions to minimize a neartight upper bound on the average loss decrease at each step while incurring negligible computational overhead We provide theoretical convergence guarantees and show that PiKE outperforms static and nonadaptive mixing baselines Furthermore we extend PiKE to promote balanced learning across tasks Extensive experiments on largescale language model pretraining confirm that PiKE achieves faster convergence and improved downstream performance compared to existing approaches |
| title | PiKE: Adaptive Data Mixing for Large-Scale Multi-Task Learning Under Low Gradient Conflicts |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2502.06244 |