TSRouter: Dynamic Modality-Model Selection for Time Series Reasoning

2026-07-09 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

TSRouter is a novel graph-based dynamic routing framework designed for time series reasoning, addressing the complementary strengths of Large Language Models (LLMs) and Vision-Language Models (VLMs). LLMs excel at exact numerical understanding but struggle with global patterns, while VLMs efficiently capture global patterns but may lose fine-grained details. TSRouter dynamically selects the most suitable modality and model for each query by constructing a heterogeneous graph that contextualizes interactions among query characteristics, modality attributes, and model capabilities. It formulates routing as a candidate scoring problem, evaluating modality-model pairs based on user-defined performance-cost preferences. Comprehensive evaluations across 4 distinct time series reasoning tasks demonstrate TSRouter's substantial performance, yielding 16% to 46% relative improvements over diverse baselines. The framework also exhibits robust zero-shot plug-and-play generalization to unseen models and novel tasks, while reducing computational overhead through cost-aware optimization.

Key takeaway

For Machine Learning Engineers building time series reasoning systems, TSRouter offers a robust approach to overcome the limitations of single-model or single-modality solutions. You should consider implementing dynamic, graph-based routing to intelligently select between text-based LLMs and vision-based VLMs, optimizing for both performance and computational cost. This framework allows your systems to adapt to diverse query characteristics, potentially yielding significant accuracy improvements and enabling zero-shot generalization to new tasks and models.

Key insights

Dynamically selecting the optimal modality and model for time series reasoning improves performance and efficiency by utilizing complementary strengths.

Principles

• LLMs preserve numerical understanding in time series.
• VLMs efficiently capture global time series patterns.
• Model selection should consider performance-cost preferences.

Method

TSRouter constructs a heterogeneous graph of task, query, modality, and model nodes, then formulates routing as a candidate scoring problem to evaluate modality-model pairs based on performance-cost preferences.

In practice

• Combine LLMs and VLMs for comprehensive time series analysis.
• Implement graph-based routing for dynamic model selection.
• Optimize model selection based on performance and cost.

Topics

• Time Series Reasoning
• Dynamic Modality Selection
• Large Language Models
• Vision-Language Models
• Graph Neural Networks
• Model Routing

Code references

• tianyi-lab/TSRouter

Best for: Research Scientist, AI Engineer, AI Scientist, Machine Learning Engineer, AI Architect

Related on AIssential

• Tyler: Typed Latent Reasoning for Language Models -- When to Think, What to Compute, and How Much to Allocate — Tyler dynamically manages latent computation in LLMs, deciding when, what type, and how much to allocate for improved reasoning.
• MSUE: Multi-Modal Soccer Understanding Expert — The paper combines VLM-driven data synthesis with an LLM-orchestrated multi-expert system for multi-modal VQA.
• Low-Latency Model Router: Automatic LLM Selection Across OpenRouter — Dynamic LLM routing optimizes model selection based on latency, cost, and quality for varied workloads.
• MODF-SIR: A Multi-agent Omni-modal Distilled Framework for Social Intelligence Reasoning — A multi-agent, omni-modal framework uses distillation and TTA with LoRA for social intelligence reasoning.
• TimeRouter: Efficient and Adaptive Routing of Time-Series Foundation Models — TimeRouter efficiently routes diverse time-series foundation models using discriminative routing and selective gating, avoiding LLM inference overhead.
• T^2MLR: Transformer with Temporal Middle-Layer Recurrence — T2MLR enhances Transformer reasoning by enabling abstract middle-layer computation to persist across decoding steps with low inference overhead.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.