Divide-and-Conquer: Tree-structured Strategy with Answer Distribution Estimator for Goal-Oriented Visual Dialogue

2025-02-09

AI Summary

Existing goal-oriented visual dialogue methods lack strategic question-asking mechanisms, resulting in stochastic search behavior and poor convergence. This paper proposes a tree-based divide-and-conquer strategy: in each dialogue turn, a binary discriminative question eliminates approximately half of the candidate objects, enabling efficient target localization. Key contributions include the first “binary reward + candidate minimization” dual-objective reinforcement learning framework, integrated with tree-structured dialogue modeling, answer distribution estimation, and dynamic candidate set pruning—ensuring both policy interpretability and guaranteed convergence. Experiments demonstrate substantial improvements in search accuracy, reduced question redundancy and dialogue turns, and generation of more discriminative questions, achieving state-of-the-art performance across multiple benchmarks.

Technology Category

• Search and Optimization: Learning to Search
• Natural Language Processing: Question Answering
• Machine Learning: Reinforcement Learning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Goal-oriented visual dialogue involves multi-round interaction between artificial agents, which has been of remarkable attention due to its wide applications. Given a visual scene, this task occurs when a Questioner asks an action-oriented question and an Answerer responds with the intent of letting the Questioner know the correct action to take. The quality of questions affects the accuracy and efficiency of the target search progress. However, existing methods lack a clear strategy to guide the generation of questions, resulting in the randomness in the search process and inconvergent results. We propose a Tree-Structured Strategy with Answer Distribution Estimator (TSADE) which guides the question generation by excluding half of the current candidate objects in each round. The above process is implemented by maximizing a binary reward inspired by the ``divide-and-conquer'' paradigm. We further design a candidate-minimization reward which encourages the model to narrow down the scope of candidate objects toward the end of the dialogue. We experimentally demonstrate that our method can enable the agents to achieve high task-oriented accuracy with fewer repeating questions and rounds compared to traditional ergodic question generation approaches. Qualitative results further show that TSADE facilitates agents to generate higher-quality questions.

Problem

• Improves question generation in visual dialogue
• Reduces randomness in search process
• Enhances accuracy with fewer dialogue rounds

Innovation

• Tree-Structured Strategy
• Answer Distribution Estimator
• Divide-and-Conquer paradigm