Researchers have developed Splice, a novel system for selecting optimal data sources for machine learning tasks. Splice aims to improve downstream task utility by identifying subsets of discovered data sources that avoid introducing noise or distribution shifts. Two faster variants, Splice-DP and Splice-Grad, were also introduced, which leverage meta-learning and gradient alignment to estimate source contributions, achieving significant speedups with comparable or improved results on classification and regression tasks. AI
IMPACT This research could lead to more efficient and accurate machine learning model training by improving data source selection.
RANK_REASON This is a research paper detailing a new system for machine learning tasks. [lever_c_demoted from research: ic=1 ai=1.0]
- Ambarish Singh
- arXiv
- CORE Recommender
- DagsHub
- Hugging Face
- IArxiv Recommender
- Splice
- Splice-DP
- Splice-Grad
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
