ZeroFolio uses text embeddings for algorithm selection without domain knowledge

Researchers have developed a novel feature-free approach to algorithm selection called ZeroFolio, which utilizes pretrained text embeddings to distinguish problem instances without requiring domain-specific knowledge. This method involves serializing the raw instance file as plain text, embedding it using a pretrained model, and then selecting an algorithm via weighted k-nearest neighbors. Evaluations on 11 ASlib scenarios demonstrated that ZeroFolio outperformed random forests trained on hand-crafted features in most cases and closely matched AutoFolio's performance without extensive tuning. AI

IMPACT This approach could enable more general-purpose algorithm selection tools that require less manual feature engineering.

RANK_REASON The cluster contains a research paper detailing a new method for algorithm selection. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• answer set programming
• ASlib
• AutoFolio
• Contrastive Semantic Projection
• MaxSAT
• Stefan Szeider
• ZeroFolio

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →