Accepted Papers
We received a strong set of submissions and are delighted to announce the 84 accepted papers below. Titles link to the corresponding OpenReview page. Best Paper and Runner-Up awards will be announced during the closing session.
Oral Presentations (5)
- Agentic Forecasting using Sequential Bayesian Updating of Linguistic Beliefs Google
- Allocation, Not Volume: Test-Time Compute for Agentic Forecasting EPFL · ELLIS Institute Tübingen / Max Planck Institute for Intelligent Systems
- Forecasting Emerges from Auto-Regressive Pretraining: Latent Predictive Structure in Language Models McGill University · Mila, Université de Montréal · University of Toronto · Concordia University · Huawei Noah's Ark Lab
- Forecasting Motion in the Wild Google DeepMind · Toyota Technological Institute at Chicago · UC Berkeley
- FutureSim: Replaying World Events to Evaluate Adaptive Agents ELLIS Institute Tübingen · Max Planck Institute for Intelligent Systems · University of Stuttgart · University of Tübingen
Spotlights (10)
- Approximate Recall, Approximate Forecasts: Recall as a Diagnostic for LLM Forecasting Errors
- Beyond Accuracy: Can LLM Forecasters Profit on Prediction Markets?
- Curating the Future: A Scalable Recipe for Training Open-Ended Forecasters
- Decentralized Aggregation of LLM Predictions via Wagering Mechanisms
- Forecast-to-Trade: Hierarchical Reinforcement Learning for Decision-Aware Financial Forecasting
- ForecastBench-Sim: A Simulated-World Forecasting Benchmark
- ForecastCompass: Guiding Agentic Forecasting with Adaptive Factor Memory
- Future-as-Label: Scalable Supervision from Real-World Outcomes
- Reaching the frontier of AI forecasting with reinforcement learning
- When do prophets profit in prediction markets?
Posters (69)
- A Black-Box Reduction from Regret to Multi-Level Coverage
- A Lightweight Deep Learning Approach to Spatiotemporal Heat Forecasting
- Accurate Forecasts Do Not Ensure Safe Decision
- AgentRx: A Benchmark for Multimodal Clinical Forecasting with LLM Agents
- Aligning LLMs with Human Uncertainty: A Beta-Bernoulli Calibrator for LLM Forecasting
- Alive and Predicting: A Live Evaluation of Multi-Step Forecasting Agents
- An adversarial tournament design for efficiently probing the frontier of AI forecasting
- Analogical Deep Research: Retrieving and Integrating Historical Analogies for Foresight Analysis
- Arbitrage-Free Forecasts from Language Models via Coherence Projection
- Auditing Actionability in AI Forecasting Interfaces
- Beyond Forecasting: The Belief-to-Trade Layer in Prediction-Market Agents
- Decision-Relevant Predictions with Joint Scoring Rules
- DELPHYNE: A Pre-Trained Model for General and Financial Time Series
- Discover then Refine: A Joint Multiple Choice Learning and Flow Matching Framework for Heat Demand Forecasting
- Diversity is the strength of the AI crowd
- Do Language Models Update their Forecasts with New Information?
- Do Time Series Foundation Model Benchmarks Hide Regime-Dependent Failures? Evidence from Traffic Speed Forecasting
- DuoMamba: A Decomposition-Free State Space Model for Long-Term Time Series Forecasting
- Efficient Forecasting of Task Failures in LLM Agents through Adaptive Fault Injection
- Elicitation Format Drives Divergent LLM Geopolitical Forecasts
- Enabling Uncertainty-Aware Time-Series Forecasting in Federated Learning for Urban Water Dynamics
- Evaluating Long-Form Forecasts by Their Effect on Downstream Predictions
- Forecasting Model Success at Inference Time: Calibrated Probabilistic Forecasts for Cost-Optimal LLM Cascades
- Forecasting Time-Varying Correlation Matrices with Large Language Models
- Forecasting With LLMs: Improved Generalization Through Feature Steering
- Forecasts as a Behavioral Probe of Language Models
- Foresight-Phys: A Benchmark for Forecasting the Results of Physical Experiments
- ForesightFlow: An Information Leakage Score Framework for Prediction Markets
- Forward-Chaining Temporal Point Process
- From Events to Impacts: Calibrated Decomposition for LLM-Based Geopolitical Forecasting
- From Marks to Narratives: Language-Augmented Spatio-Temporal Point Processes
- From Narrative to Auditable Forecasts: An Agentic Scaffold for Probabilistic Forecasting
- Generative Bayesian Computation for Probabilistic Forecasting with Discrete Events
- GENERATIVE TRAFFIC FORECASTING: PRESERVING SHOCKWAVE TOPOLOGY WITH DIFFUSION MODELS
- HMTMO-GP: Hierarchical Multi-Task Multi-Output Gaussian Processes
- How Predictable is AI Progress?
- Iterative Computation as Anytime Forecasting: Dense Supervision for Calibrated Trajectories in Recurrent World Models
- Latent Market Dynamics: A World Model Framework for Agentic Prediction Markets
- Latent Stochastic Interpolants for Probabilistic Time Series Forecasting
- Leakage-Aware Benchmarking of LLM Forecasting: Real-Time Nowcasts as the Decision-Time Input for Macro Factor Ranking
- MacroBench: Measuring Frontier LLM Macroeconomic Forecasting Ability
- Measuring Source-Induced Bias in LLM Forecasts with Prediction Markets
- Mix, Don’t Pick: Why Synthetic Corpus Composition Matters for Time Series Foundation Model Pretraining
- One Token per Trade: Multi-Resolution Limit Order Book Forecasting with a Foundation Model
- OptimismBench: Measuring Forecasting Bias in Language Model Judgment
- Outcome-Free Audits and Repairs for LLM Forecasters
- Period-Aware Inductive Bias Versus Scale on Influenza-like Illness Forecasting
- Physics-Informed Bidirectional Graph Networks for Traffic Prediction: Deriving Message Passing Direction from Traffic Flow Theory
- Polynomial Input Preconditioning for Zero-Shot Time Series Forecasting
- Preference Optimization Drives Monoculture in LLM Prediction Markets
- Presentation Robustness for LLM Forecasters
- Proxy Scoring Enables Benchmarking LLM Forecasters Without Waiting for Outcomes
- Quantizing Time-Series Models As Dynamical Systems: Trajectory-Based Quantization Sensitivity Score
- Reflexivity as Prompt: Does Awareness of Self-Reinforcing Market Dynamics Improve LLMs as Financial Market Forecasters?
- Repackaging Temporal Evidence: A Unifying Interface for Temporal Prediction
- RIFT: Reliability of LLM and Physics Forecasters across Time-Horizons in Coupled PDE Systems
- Robustness of Multimodal Foundation-Model Forecasting for Postoperative Cancer Outcomes
- SC-JEPA: Stabilizing Latent Predictive Learning for Time-Series Anomaly Prediction
- SciPaths: Forecasting Pathways to Scientific Discovery
- Semantics-Enhanced Retrieval-Augmented Time Series Forecasting
- Simulation-Augmented Multi-Step Split Conformal Prediction for Aggregated Forecasts
- StretchTime: Adaptive Time Series Forecasting via Symplectic Attention
- Temporally Supervised Linear Probes Improve LLM Forecasts
- TimeRouter: Efficient and Adaptive Routing of Time Series Foundation Models
- VSTF: Vision and Sequence Models for Time-Frequency Time Series Forecasting
- What if Tomorrow is the World Cup Final? Counterfactual Time Series Forecasting with Textual Conditions
- What Should We Forecast? Benchmarking Agents on Early Question Discovery
- When Does Evidence Help Prompted LLM Forecasting? Evidence Access and Prompt Structure Across 12 Models
- WorldFork: Auditable Branching Rollouts for LLM Forecasting
Call for Papers
We invite submissions on all aspects of AI forecasting, from methodological advances to benchmark design to applications in real-world domains. Papers should be submitted via OpenReview and will undergo peer review by our program committee.
Accepted papers will be presented as posters during the workshop, with selected papers invited for oral presentations. A best paper award will be given at the closing session.
Key Dates
| Event | Date |
|---|---|
| Submission Portal Opens | April 21, 2026 |
| Abstract Registration Deadline | May 11, 2026 (11:59 PM UTC, not AoE) |
| Submission Deadline | May 13, 2026 (11:59 PM UTC, not AoE) |
| Reviewer Bidding | May 15–18, 2026 |
| Review Period | May 20 – June 8, 2026 |
| Author Notification | June 10, 2026 |
Submission Guidelines
Format: Submissions should be up to 4 pages (excluding references and appendix) using the ICML 2026 template.
Anonymity: All submissions should be anonymized for double-blind review.
Non-archival & Dual Submission: The workshop is non-archival, so dual submission is allowed — we welcome submissions of work that has been previously published or is under review elsewhere, with proper disclosure.
Platform: Submissions will be handled through OpenReview.
Forecasting Agent Hackathon
Ahead of the workshop, we co-hosted the AI Forecasting Hackathon (May 16–17, 2026), where participants built forecasting agents and competed on the Prophet Arena leaderboard.
Congratulations to our winners, Hanson Wen (UC Berkeley) and James Gui (USC), who will share their approach during the Hackathon Winner Presentation in the workshop program, and to our runner-up, Shirish Chinchanikar (UChicago).
FAQ
How do I submit a camera-ready version of my paper?
Go back into OpenReview, open your submission, click Edit, and feel free to upload an edited version of your paper. Please make any edits by June 25, 2026, 11:59 pm UTC.
What format should my poster follow?
Posters should follow the ICML 2026 poster instructions, though there is some flexibility in the format.
What does a spotlight mean?
A spotlight is presented as a poster, following the same format described above. In addition, we will place a sign next to your poster indicating that your submission achieved a higher bar.
What does an oral mean?
An oral is an 8–10 minute presentation given at the workshop explaining your approach and results. We strongly prefer in-person presentations — please email us ASAP if you won't be able to attend.
Please prepare a set of slides and email them to us at forecastworkshop@gmail.com, and let us know if you need any special arrangements.
When will our presentation time be?
We will allocate presentation times later — please email us if you have preferences. The slots are listed on this website under Schedule.
Which ICML registration should I buy?
We recommend attending the full conference, but if you aren't interested in the main conference, feel free to just buy the workshop registration. Please note that we do not reimburse registration fees.
More to come — we'll keep adding answers here as more questions arrive. Have a question? Email us at forecastworkshop@gmail.com.