Time Series in the Age of Large Models

Workshop at the International Conference on Learning Representations (ICLR) 2026

The 1st ICLR Workshop on Time Series in the Age of Large Models (TSALM) will be held at the Riocentro Convention and Event Center in Rio de Janeiro, Brazil on April 26, 2026. We look forward to welcoming you in Rio de Janeiro.

Introduction

Building on our NeurIPS 2024 workshop, this iteration focuses on key open problems in time series foundation models: context-informed predictions, reasoning and agents, interpretability, and rigorous evaluation.

While recent time series foundation models have shown promise for univariate forecasting, most practical scenarios require additional context—multivariate relationships, exogenous variables, or information from other modalities such as text. Simultaneously, LLM-powered agents are opening new possibilities for automating time series analysis workflows, though research in this area remains nascent. As these models are increasingly deployed in high-stakes domains, understanding their inner workings and failure modes becomes critical.

The key topics of this workshop include:

  • Context-informed foundation models: Multivariate forecasting, static and dynamic exogenous variables, multimodal time series, anomaly detection, and classification
  • Reasoning and agents: LLM-powered agents for time series tasks, agentic scaffolds, and benchmarks for evaluating agent performance
  • Benchmarks, datasets, and tools: Unified evaluation frameworks, large-scale and live benchmarks, synthetic data generation, and open-source libraries for pretraining and inference
  • Interpretability: Attention analysis, feature attribution, symbolic distillation, and causal interpretability for large time series models
  • Evaluation and applications: Metrics for probabilistic prediction, domain-specific evaluation, and applications in finance, healthcare, climate science, and dynamical systems

Please see the Call for Papers for details.

We are very grateful to have been sponsored by Susquehanna.

SIGlogo

Thanks to our sponsor, there will be cash awards for the best papers!
We will also have exclusive snacks and refreshments directly in the workshop hall.
Join us if you'll be attending in person!

Schedule

April 26, 2026, Riocentro Convention and Event Center, Rio de Janeiro, Brazil

All times are in the America/Sao_Paulo time zone (UTC-3). The schedule will be further refined in the weeks leading up to the workshop.

Time Event
09:00 - 09:10 🎤 Opening Remarks
09:10 - 09:45 🎓 Invited Talk by Bernie Wang
09:45 - 10:00 💡 Spotlight Session 1

09:45 - 09:50: Hindsight Preference Optimization for Financial Time Series Advisory
09:50 - 09:55: TS-Arena: A Live Forecast Pre-Registration Platform for Leakage-Free Evaluation of Time Series Foundation Models
09:55 - 10:00: Narrative of Time across Scales (NoTS)
10:00 - 10:25 Coffee Break
10:25 - 11:00 🎓 Invited Talk by Pablo Montero-Manso
11:00 - 12:00 🖼️ Poster Session 1
12:00 - 13:00 🥗 Lunch Break
13:00 - 13:30 🤝 Dedicated Networking Session
13:30 - 13:55 💡 Spotlight Session 2

13:30 - 13:35: ARFBench: Benchmarking Multimodal Time Series Reasoning for Software Incident Response
13:35 - 13:40: Overcoming the Modality Gap in Context-Aided Forecasting
13:40 - 13:45: Finding the Zeitgeist in Time Series Foundation Models
13:45 - 13:50: The Arrow of Time: What Tabular Foundation Models Miss in Time Series Forecasting
13:50 - 13:55: Does Normalization Choice Matter for Causal Large Time-Series Models?
13:55 - 15:00 🖼️ Poster Session 2
15:00 - 15:30 Coffee Break
15:30 - 16:05 🎓 Invited Talk by Rose Yu
16:05 - 16:40 🎓 Invited Talk by Othmane Abou-Amal
16:40 - 17:00 🎬 Closing Remarks