Evolution of Thought: Diverse and High-Quality Reasoning via Multi-Objective Optimization

TL;DR

Evolution of Thought (EoT) reframes multi-modal reasoning as a multi-objective optimization problem that jointly maximizes answer quality and diversity. It leverages NSGA-II with crossover/mutation to explore diverse high-quality candidates and introduces a condensation-aggregation (CA) mechanism to distill information and produce accurate final answers. Across MathVista, Math-Vision, and GSM8K, EoT demonstrates superior diversity, competitive or improved quality, and greater efficiency compared with strong baselines. This framework offers a principled, heuristic approach to robust, scalable reasoning for multi-modal large language models.

Abstract

As multi-modal large language models (MLLMs) are increasingly applied to complex reasoning tasks, the diversity and quality of reasoning paths become crucial factors affecting their performance. Although current methods aim to enhance reasoning quality through path expansion, they often neglect the diversity of reasoning paths and effective information sharing, leading to local optima and inefficiency. To address these challenges, we propose Evolution of Thought (EoT), a multi-objective framework designed to improve reasoning by fostering both high-quality and diverse reasoning paths. Specifically, we introduce the Non-dominated Sorting Genetic Algorithm II for multi-objective optimization, utilizing crossover and mutation operators to promote greater diversity in reasoning solutions. Additionally, we propose a Condensation-Aggregation mechanism to cluster and eliminate redundant paths, facilitate improved information sharing among parent nodes, and ultimately enhance both the efficiency and quality of the reasoning process. Validation experiments on various vision-language and language reasoning tasks demonstrate that EoT achieves superior reasoning performance and efficiency compared to other competitive baselines. Our study provides a novel perspective on the design of heuristic reasoning frameworks for MLLMs.

Figures (10)

• Figure 1: Comparison of Evolution of Thoughts (EoT) with other approaches. Color depth indicates answer quality: blue for high quality, red for low quality.
• Figure 2: EoT Process Flow Diagram.Step 1 involves scoring using $\mathcal{M}^{Q}(A)$ and $\mathcal{M}^{N}(A)$; Step 2 involves non-dominated sorting; Step 3 includes Crossover and Mutation operations to generate diverse offspring.
• Figure 3: Abalation results on different $N,T,r,K,\kappa,m$ parameters and datsets. Fig (a) shows pass@4 accuracy (%) on different candidate answers $N$ and evolutionary generations $T$. Fig (b) and (c) show the pass@$k$ ($k=1,2,...,12$) accuracy (%) curves under different crossover-to-mutation ratios $r$ and selected candidate parents $K$. Fig (d) shows the average accuracy on three datasets under different clusters $\kappa$ and drop clusters $m$.
• Figure :
• Figure :