TSRouter: Dynamic Modality-Model Selection for Time Series Reasoning
2026-07-09 • Machine Learning
Machine Learning
AI summaryⓘ
The authors address how different AI models understand time series data, which is important for solving real-world problems. They note that language-based models keep exact numbers but miss big-picture patterns, while vision-language models spot these patterns but lose some details. To balance this, the authors created TSRouter, a system that smartly picks the best model type for each task by analyzing how tasks, data, and models interact. Their tests show TSRouter works better than other methods, adapts well to new tasks or models, and saves computing resources.
time series reasoningLarge Language Models (LLMs)Vision-Language Models (VLMs)heterogeneous graphdynamic routingmodel selectionzero-shot generalizationperformance-cost tradeoffgraph-based frameworktime series tasks
Authors
Fangxu Yu, Tao Feng, Dehai Min, Lu Cheng, Ge Liu, Tianyi Zhou
Abstract
Time series reasoning is essential for real-world problem-solving. While both Large Language Models (LLMs) and Vision-Language Models (VLMs) can reason about time-series data, their capabilities are complementary: LLMs process time series as text sequences and thus preserve exact numerical understanding, but struggle with global patterns, whereas VLMs efficiently capture these patterns by visualizing time series but may lose fine-grained details. Moreover, models vary significantly in task-specific expertise and inference costs. Dynamically selecting the most suitable modality and model for each query is therefore crucial, yet challenging because it requires modeling the complex interactions among tasks, queries, modalities, and models, which carry rich contextual signals. To this end, we introduce TSRouter, a graph-based dynamic routing framework. TSRouter constructs a heterogeneous graph of task, query, modality, and model nodes to contextualize the interactions among query characteristics, modality attributes, and model capabilities. TSRouter formulates routing as a candidate scoring problem, where each modality-model pair is evaluated based on user-defined performance-cost preferences to select the optimal candidate. Comprehensive evaluations on 4 distinct time series reasoning tasks reveal that TSRouter substantially outperforms diverse baselines with 16\% to 46\% relative improvements. Furthermore, TSRouter demonstrates robust zero-shot plug-and-play generalization to unseen models and novel tasks and preserves high performance while reducing computational overhead through cost-aware optimization. Our code is available at https://github.com/tianyi-lab/TSRouter.