RZ-NAS: Enhancing LLM-guided Neural Architecture Search via Reflective Zero-Cost Strategy

Zipeng Ji, Guanghui Zhu, Chunfeng Yuan, Yihua Huang

TL;DR: A novel framework that combines the text- and code-level comprehension capabilities of LLMs with a Reflective Zero-Cost evaluation strategy for NAS

Abstract: LLM-to-NAS is a promising field at the intersection of Large Language Models (LLMs) and Neural Architecture Search (NAS), as recent research has explored the potential of architecture generation leveraging LLMs on multiple search spaces. However, the existing LLM-to-NAS methods face the challenges of limited search spaces, time-cost search efficiency, and uncompetitive performance across standard NAS benchmarks and multiple downstream tasks. In this work, we propose the Reflective Zero-cost NAS (RZ-NAS) method that can search NAS architectures with humanoid reflections and training-free metrics to elicit the power of LLMs. We rethink LLMs’ roles in NAS in current work and design a structured, prompt-based to comprehensively understand the search tasks and architectures from both text and code levels. By integrating LLM reflection modules, we use LLM-generated feedback to provide linguistic guidance within architecture optimization. RZ-NAS enables effective search within both micro and macro search spaces without extensive time cost, achieving SOTA performance across multiple downstream tasks.

Lay Summary: We introduce a novel framework that combines the text- and code-level comprehension capabilities of LLMs with a Reflective Zero-Cost evaluation strategy for neural architecture search (NAS). To integrate the text- and code-level understanding abilities of LLMs, we develop structured prompts to precisely define NAS tasks. These prompts include: a high-level role, detailed instructions, an in-context example, and the key reflective module. Moreover, we utilize Zero-Cost proxies instead of training architectures to reduce computational resources and time cost while maintaining competitive performance. The reflective module guides the LLM to reflect on mutation performance and generates targeted suggestions for further iteration improvements.

Link To Code: https://github.com/PasaLab/RZ-NAS

Primary Area: Deep Learning->Algorithms

Keywords: Automated Machine Learning, Neural Architecture Search, Large Language Models

Submission Number: 14855