Language-Guided Tuning: Enhancing Numeric Optimization with Textual Feedback

2025-08-21

AI Summary

Machine learning configuration optimization suffers from tightly coupled multidimensional design spaces (e.g., architecture, training strategies, hyperparameters), poor interpretability, and limited adaptability to dynamic conditions. To address these challenges, we propose the Language-Guided Tuning (LGT) framework—a multi-agent closed-loop system comprising Consultant, Evaluator, and Optimizer agents. LGT is the first approach to integrate natural language inference and textual gradients—qualitative, human-interpretable feedback signals—into configuration search, enabling semantic-level collaborative tuning and self-improvement. The method synergistically combines large language models, textual gradient analysis, and numerical optimization to support configuration suggestion generation, progress assessment, and decision refinement. Evaluated on six benchmark datasets, LGT significantly outperforms conventional optimization methods while delivering high interpretability and the capacity to model semantic relationships within complex, high-dimensional configuration spaces.

Technology Category

• Natural Language Processing: Learning & Optimization for NLP
• Search and Optimization: Learning to Search
• Machine Learning: Optimization

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Configuration optimization remains a critical bottleneck in machine learning, requiring coordinated tuning across model architecture, training strategy, feature engineering, and hyperparameters. Traditional approaches treat these dimensions independently and lack interpretability, while recent automated methods struggle with dynamic adaptability and semantic reasoning about optimization decisions. We introduce Language-Guided Tuning (LGT), a novel framework that employs multi-agent Large Language Models to intelligently optimize configurations through natural language reasoning. We apply textual gradients - qualitative feedback signals that complement numerical optimization by providing semantic understanding of training dynamics and configuration interdependencies. LGT coordinates three specialized agents: an Advisor that proposes configuration changes, an Evaluator that assesses progress, and an Optimizer that refines the decision-making process, creating a self-improving feedback loop. Through comprehensive evaluation on six diverse datasets, LGT demonstrates substantial improvements over traditional optimization methods, achieving performance gains while maintaining high interpretability.

Problem

• Optimizing interdependent ML configurations lacking interpretability and adaptability
• Integrating semantic reasoning with numerical optimization through textual feedback
• Coordinating multi-agent language models for intelligent configuration tuning decisions

Innovation

• Multi-agent LLMs for natural language optimization
• Textual gradients provide semantic feedback signals
• Coordinated advisor-evaluator-optimizer feedback loop