An approach for systematic decomposition of complex llm tasks

2025-10-09

AI Summary

To address the reliability deficits of large language models (LLMs) on complex tasks—stemming from heuristic or manual task decomposition—this paper proposes ACONIC, the first systematic task decomposition framework grounded in constraint modeling and formal complexity analysis. Methodologically, ACONIC introduces constraint-induced complexity analysis: it formalizes tasks as constraint satisfaction problems and leverages computational complexity metrics to automatically determine optimal decomposition granularity and branching paths. Crucially, it eliminates reliance on agent orchestration or human priors, thereby enhancing decomposition stability and interpretability. Evaluated on SATBench (combinatorial optimization) and Spider (text-to-SQL), ACONIC improves LLM accuracy by 10–40 percentage points, demonstrating both effectiveness in complex reasoning and strong generalization across diverse domains.

Technology Category

• Constraint Satisfaction and Optimization: Satisfiability Modulo Theories
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• 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

Abstract

Large Language Models (LLMs) suffer from reliability issues on complex tasks, as existing decomposition methods are heuristic and rely on agent or manual decomposition. This work introduces a novel, systematic decomposition framework that we call Analysis of CONstraint-Induced Complexity (ACONIC), which models the task as a constraint problem and leveraging formal complexity measures to guide decomposition. On combinatorial (SATBench) and LLM database querying tasks (Spider), we find that by decomposing the tasks following the measure of complexity, agent can perform considerably better (10-40 percentage point).

Problem

• Systematic decomposition of complex LLM tasks
• Addressing reliability issues in LLM performance
• Modeling tasks as constraint problems for decomposition

Innovation

• Systematic decomposition framework for complex LLM tasks
• Models tasks as constraint problems with formal complexity
• Guides decomposition using formal complexity measures