Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DeALOG: Decentralized Multi-Agents Log-Mediated Reasoning Framework

Abhijit Chakraborty, Ashish Raj Shekhar, Shiven Agarwal, Vivek Gupta

TL;DR

DeALOG introduces a planner-free, decentralized multi-agent framework for complex multimodal question answering. Five specialized agents collaborate via a persistent natural-language log to produce, verify, and repair evidence-based answers, enabling transparent intermediate reasoning and robust long-horizon reasoning. Across six challenging benchmarks, DeALOG achieves competitive or state-of-the-art accuracy, with ablations showing the shared log and verification are crucial for faithfulness and error resilience. The work demonstrates that log-mediated, peer-verified coordination can yield scalable, robust multimodal QA without task-specific fine-tuning.

Abstract

Complex question answering across text, tables and images requires integrating diverse information sources. A framework supporting specialized processing with coordination and interpretability is needed. We introduce DeALOG, a decentralized multi-agent framework for multimodal question answering. It uses specialized agents: Table, Context, Visual, Summarizing and Verification, that communicate through a shared natural-language log as persistent memory. This log-based approach enables collaborative error detection and verification without central control, improving robustness. Evaluations on FinQA, TAT-QA, CRT-QA, WikiTableQuestions, FeTaQA, and MultiModalQA show competitive performance. Analysis confirms the importance of the shared log, agent specialization, and verification for accuracy. DeALOG, provides a scalable approach through modular components using natural-language communication.

DeALOG: Decentralized Multi-Agents Log-Mediated Reasoning Framework

TL;DR

DeALOG introduces a planner-free, decentralized multi-agent framework for complex multimodal question answering. Five specialized agents collaborate via a persistent natural-language log to produce, verify, and repair evidence-based answers, enabling transparent intermediate reasoning and robust long-horizon reasoning. Across six challenging benchmarks, DeALOG achieves competitive or state-of-the-art accuracy, with ablations showing the shared log and verification are crucial for faithfulness and error resilience. The work demonstrates that log-mediated, peer-verified coordination can yield scalable, robust multimodal QA without task-specific fine-tuning.

Abstract

Complex question answering across text, tables and images requires integrating diverse information sources. A framework supporting specialized processing with coordination and interpretability is needed. We introduce DeALOG, a decentralized multi-agent framework for multimodal question answering. It uses specialized agents: Table, Context, Visual, Summarizing and Verification, that communicate through a shared natural-language log as persistent memory. This log-based approach enables collaborative error detection and verification without central control, improving robustness. Evaluations on FinQA, TAT-QA, CRT-QA, WikiTableQuestions, FeTaQA, and MultiModalQA show competitive performance. Analysis confirms the importance of the shared log, agent specialization, and verification for accuracy. DeALOG, provides a scalable approach through modular components using natural-language communication.
Paper Structure (52 sections, 5 figures, 10 tables, 1 algorithm)

This paper contains 52 sections, 5 figures, 10 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. De alog Framework
  3. Design goals and assumptions.
  4. Agent roles and I/O contracts.
  5. Shared log and types.
  6. Controller loop and stopping.
  7. Verification and one‑shot re‑engagement.
  8. Memory management and complexity.
  9. Experiments
  10. Input Filtering and Retrieval:
  11. Evaluation Approach:
  12. LLM Models:
  13. Baseline Methods:
  14. Results and Analysis
  15. Why we underperformed in TATQA:
  16. ...and 37 more sections

Figures (5)

  • Figure 1: Example of multi-hop table question answering. The system identifies the author of "Eat Pray Love", extracts metadata from a passage and a table, reasons over birth years and nationalities to find the correct answer.
  • Figure 2: De alog: Planner‑free, log‑mediated QA. Agents read/write to a shared log; the Summarizer synthesizes, the Verifier cross‑checks.
  • Figure 3: Latency comparison across methods shows that De alog has slightly higher average response time per query due to multi-agent coordination.
  • Figure 4: Catastrophic error rates under increasing corruption levels comparing Planner, Plan→Log hybrid, and De alog (top image). Exact match (EM) comparison of robustness and long-horizon reasoning performance between a re-planning Planner, a Plan→Log hybrid, and De alog (bottom table).
  • Figure 5: Revenue by year extracted by the VisualAgent (2020: $5.2M, 2021: $6.1M).