Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TS-Debate: Multimodal Collaborative Debate for Zero-Shot Time Series Reasoning

Patara Trirat, Jin Myung Kwak, Jay Heo, Heejun Lee, Sung Ju Hwang

TL;DR

TS-Debate introduces a cross-modal, collaborative agent framework for zero-shot time-series reasoning by assigning modality-specific Text, Visual, and Numerical analysts, anchored by a domain knowledge elicitation stage and a verification-conflict-calibration protocol. The framework preserves modality fidelity, surfaces conflicting evidence, and mitigates numeric hallucinations without task-specific training. It coordinates analysis through iterative rounds, code execution, and numerical lookups to verify claims against data and domain constraints. Across 20 tasks on MTBench, TimerBed, and TSQA, TS-Debate achieves consistent improvements over strong baselines, demonstrating the value of explicit verification and structured cross-modal debate for robust time-series reasoning.

Abstract

Recent progress at the intersection of large language models (LLMs) and time series (TS) analysis has revealed both promise and fragility. While LLMs can reason over temporal structure given carefully engineered context, they often struggle with numeric fidelity, modality interference, and principled cross-modal integration. We present TS-Debate, a modality-specialized, collaborative multi-agent debate framework for zero-shot time series reasoning. TS-Debate assigns dedicated expert agents to textual context, visual patterns, and numerical signals, preceded by explicit domain knowledge elicitation, and coordinates their interaction via a structured debate protocol. Reviewer agents evaluate agent claims using a verification-conflict-calibration mechanism, supported by lightweight code execution and numerical lookup for programmatic verification. This architecture preserves modality fidelity, exposes conflicting evidence, and mitigates numeric hallucinations without task-specific fine-tuning. Across 20 tasks spanning three public benchmarks, TS-Debate achieves consistent and significant performance improvements over strong baselines, including standard multimodal debate in which all agents observe all inputs.

TS-Debate: Multimodal Collaborative Debate for Zero-Shot Time Series Reasoning

TL;DR

TS-Debate introduces a cross-modal, collaborative agent framework for zero-shot time-series reasoning by assigning modality-specific Text, Visual, and Numerical analysts, anchored by a domain knowledge elicitation stage and a verification-conflict-calibration protocol. The framework preserves modality fidelity, surfaces conflicting evidence, and mitigates numeric hallucinations without task-specific training. It coordinates analysis through iterative rounds, code execution, and numerical lookups to verify claims against data and domain constraints. Across 20 tasks on MTBench, TimerBed, and TSQA, TS-Debate achieves consistent improvements over strong baselines, demonstrating the value of explicit verification and structured cross-modal debate for robust time-series reasoning.

Abstract

Recent progress at the intersection of large language models (LLMs) and time series (TS) analysis has revealed both promise and fragility. While LLMs can reason over temporal structure given carefully engineered context, they often struggle with numeric fidelity, modality interference, and principled cross-modal integration. We present TS-Debate, a modality-specialized, collaborative multi-agent debate framework for zero-shot time series reasoning. TS-Debate assigns dedicated expert agents to textual context, visual patterns, and numerical signals, preceded by explicit domain knowledge elicitation, and coordinates their interaction via a structured debate protocol. Reviewer agents evaluate agent claims using a verification-conflict-calibration mechanism, supported by lightweight code execution and numerical lookup for programmatic verification. This architecture preserves modality fidelity, exposes conflicting evidence, and mitigates numeric hallucinations without task-specific fine-tuning. Across 20 tasks spanning three public benchmarks, TS-Debate achieves consistent and significant performance improvements over strong baselines, including standard multimodal debate in which all agents observe all inputs.
Paper Structure (57 sections, 2 equations, 4 figures, 9 tables, 1 algorithm)

This paper contains 57 sections, 2 equations, 4 figures, 9 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Proposed Framework: TS-Debate
  4. Problem Setting
  5. Framework Overview
  6. Multimodal Query Construction
  7. Multimodal Collaborative Debate
  8. Verification-Conflict-Calibration (VCC) Protocol
  9. Experiments
  10. Setup
  11. Main Results
  12. Ablation and Sensitivity Analysis
  13. Multimodal Debate Agents
  14. Component Variations
  15. Debate Configurations
  16. ...and 42 more sections

Figures (4)

  • Figure 1: Overview of multimodal time-series reasoning: (a) complementary strengths and failure modes across modalities; (b) limitations of existing zero-shot approaches; and (c) TS-Debate, a cross-modal agentic framework with explicit verification.
  • Figure 2: Overview of the TS-Debate framework. (a) Multimodal query construction derives textual, numerical, and visual representations from user queries, raw time series, and contextual information. (b) Multimodal collaborative debate enables numerical, visual, and text analysts to reason over these representations and produce structured observations, inferences, and limitations. (c) VCC protocol applies tool-assisted review to assess debate outputs, identify conflicts, and synthesize the final answer.
  • Figure 3: Analysis results. Impact of (a) modality composition, (b) system components, and (c--e) debate configurations across tasks.
  • Figure 4: Average time and monetary cost breakdown.