[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82104-en":3,"doc-seo-82104-105":30,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},82104,1374391975076,"Riley","https://ap-avatar.wpscdn.com/avatar/14000253ca4ec9f6853?x-image-process=image/resize,m_fixed,w_180,h_180&k=1783305029341752051",8,"Research & Report","TSROUTER: Dynamic Modality-Model Selection for Time Series Reasoning","Time series reasoning is critical for real-world decision support. Existing approaches using LLMs and VLMs are complementary yet inconsistent: LLMs retain numerical precision but weaken on global patterns and suffer from context constraints, while VLMs capture global trends through visualization but can lose fine-grained detail. Different models also trade off task expertise and inference cost. TSROUTER introduces graph-based dynamic routing to score modality–model candidates under performance–cost preferences, improving 4 reasoning tasks by 16%–46% while enabling zero-shot plug-and-play generalization with lower computation overhead.","arXiv :2607 .08940v 1 [ cs .LG] 9 Jul 2026  \n TSROUTER: Dynamic Modality-Model Selection for Time Series Reasoning  \nFangxu Yu1, Tao Feng2, Dehai Min3, Lu Cheng3, Ge Liu2, Tianyi Zhou4  \n1University of Maryland, College Park, 2University of Illinois Urbana-Champaign, 3University of Illinois Chicago, 4MBZUAI  \nAbstract  \nTime 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 taskspecific 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.  \nOur code is available at [https://github.com/tianyi-lab/TSRouter](https://github.com/tianyi-lab/TSRouter).  \n1 Introduction  \nTime series plays a critical role in high-stakes domains where time series records the temporal evolution of real-world variables. Interpreting these sequential patterns is fundamental to understanding complex processes or guiding proactive decisions. For example, in education (Mao et al., 2024) , analyzing student trajectories such as engagement and attempt counts helps identify struggling learners and personalize study plans. In scientific discovery (Yu et al., 2025c), time series such as gravitational wave recordings encode astrophysical processes that may lead to discoveries of exoplanets and black-hole mergers. Analyzing time series and conducting reasoning enable automated systems to support high-impact applications.  \nRecently, Large Language Models (LLMs) and Vision-Language Models (VLMs) have emerged as powerful tools for time series analysis. LLM-based methods (Merrill et al., 2024; Wang et al., 2025) treat time series as raw text sequences, preserving numerical precision but suffering from context length bottlenecks, high computational costs, and poor global pattern recognition. Conversely, VLM-based approaches (Zhong et al., 2025; Liu et al., 2025) convert time series into visual plots, efficiently capturing global patterns in long sequences but losing fine-grained numerical detail due to image resolution limits. Beyond modality choice, individual models also vary in their ability to handle different types  \n\n| Method | Modality\u003Cbr>Routing | Interaction |  | Efficiency |  |\n| --- | --- | --- | --- | --- | --- |\n|  |  | Query-Query | Query-Model | Inference Speed | Memory Use |\n| CausalLM (Ding et al., 2024a) | ✗ | ✗ | ✗ | Slow | Medium |\n| GraphRouter (Feng et al., 2024) | ✗ | ✗ | ✓ | Fast | Low |\n| Hybrid LLM (Hu et al., 2024) | ✗ | ✗ | ✗ | Medium | Medium |\n| RouterDC (Chen et al., 2024b) | ✗ | ✓ | ✗ | Medium | Medium |\n| Router-R1 (Zhang et al., 2025a) | ✗ | ✗ | ✗ | Medium |","cbCailAHlW4L9zTI","https://ap.wps.com/l/cbCailAHlW4L9zTI","pdf",1281202,2,1,21,"English","en",105,"# Introduction\n## Background and Motivation\n## Limitations of Existing Routing Approaches\n## Proposed Method: TSROUTER","[{\"question\":\"Why do LLMs and VLMs have complementary strengths for time series reasoning?\",\"answer\":\"LLMs process time series as textual sequences, preserving numerical precision but struggling with global patterns and facing context and compute constraints. VLMs visualize time series to capture global patterns effectively, while potentially losing fine-grained numerical detail due to visual resolution limits.\"},{\"question\":\"What problem does TSROUTER address that earlier routing methods miss?\",\"answer\":\"TSROUTER jointly performs modality routing and model routing, while prior methods mostly focus on selecting models within a fixed modality. It also models complex interactions among tasks, queries, modalities, and models instead of capturing only one interaction type.\"},{\"question\":\"How does TSROUTER choose the best modality-model pair for each query?\",\"answer\":\"TSROUTER builds a heterogeneous graph over task, query, modality, and model nodes, then formulates routing as a candidate scoring problem. Each modality–model pair is scored using user-defined performance–cost preferences to select the optimal candidate.\"}]",1784178232,53,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"tsrouter-dynamic-modality-model-selection-for-time-series-reasoning","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,47,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":20},"https://docshare.wps.com/document/","Document",{"item":48,"name":12,"@type":43,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/tsrouter-dynamic-modality-model-selection-for-time-series-reasoning/82104/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-22","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"Why do LLMs and VLMs have complementary strengths for time series reasoning?","Question",{"text":75,"@type":76},"LLMs process time series as textual sequences, preserving numerical precision but struggling with global patterns and facing context and compute constraints. VLMs visualize time series to capture global patterns effectively, while potentially losing fine-grained numerical detail due to visual resolution limits.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What problem does TSROUTER address that earlier routing methods miss?",{"text":80,"@type":76},"TSROUTER jointly performs modality routing and model routing, while prior methods mostly focus on selecting models within a fixed modality. It also models complex interactions among tasks, queries, modalities, and models instead of capturing only one interaction type.",{"name":82,"@type":73,"acceptedAnswer":83},"How does TSROUTER choose the best modality-model pair for each query?",{"text":84,"@type":76},"TSROUTER builds a heterogeneous graph over task, query, modality, and model nodes, then formulates routing as a candidate scoring problem. Each modality–model pair is scored using user-defined performance–cost preferences to select the optimal candidate.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]