PulseAugur
EN
LIVE 07:29:56

New theory models LLM use in neural architecture search

By PulseAugur Editorial · [2 sources] ·

Researchers have developed a theoretical framework for understanding how large language models (LLMs) can be used in iterative neural architecture search (NAS). The proposed parametric Cross-Entropy method models LLM-NAS as an optimization process over executable programs, proving that architecture quality increases monotonically and that elite-set probability converges geometrically. The study also introduces a delta-based generation technique for higher valid-generation rates and a MinHash-Jaccard novelty filter to prevent mode collapse. Furthermore, it establishes a closed-form solution for proxy reliability, identifying a key condition for trustworthy proxy-based rankings. AI

IMPACT Provides a theoretical foundation for using LLMs in automated architecture design, potentially accelerating AI model development.

RANK_REASON The cluster contains an academic paper detailing a new theoretical framework for LLM-based neural architecture search.

Read on arXiv cs.LG →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

New theory models LLM use in neural architecture search

COVERAGE [2]

  1. arXiv cs.LG TIER_1 English(EN) · Santosh Premi Adhikari, Radu Timofte, Dmitry Ignatov ·

    Convergence Theory for Iterative LLM-Based Neural Architecture Search: A Parametric Cross-Entropy Framework with Closed-Form Proxy Reliability

    arXiv:2605.30103v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly used as generators in iterative neural architecture search (NAS), yet no formal convergence theory exists for this class of algorithms. We model iterative LLM-NAS as a parametric Cross-E…

  2. arXiv cs.LG TIER_1 English(EN) · Dmitry Ignatov ·

    Convergence Theory for Iterative LLM-Based Neural Architecture Search: A Parametric Cross-Entropy Framework with Closed-Form Proxy Reliability

    Large language models (LLMs) are increasingly used as generators in iterative neural architecture search (NAS), yet no formal convergence theory exists for this class of algorithms. We model iterative LLM-NAS as a parametric Cross-Entropy (CE) method over executable programs and …

RELATED ENTITIES

RELATED TOPICS