NeurIPS 2026 Workshop · Sydney
Call for Papers
We invite work on foundation models for temporal systems: forecasting, simulation, multimodal environments, and reliable decision-making under real-world complexity.
- Deadline
- August 29, 2026
11:59 pm AoE
- Submission
- Up to 4 pages
Double-blind · non-archival
- Review
- Three reviews
OpenReview · no reviewer LLM use
Timeline
Important dates
-
Submission deadline
August 29, 202611:59 pm AoE
-
Author notification
September 29, 2026 Fixed
-
Camera-ready
November 6, 2026Tentative
-
Workshop
December 11–12, 2026Exact day and room TBA
The notification date is set by the NeurIPS workshop chairs and cannot be extended. We therefore cannot grant submission-deadline extensions that would compress the review period.
Submission format
Submissions
A single track. Submissions may be up to 4 pages, excluding references and appendices; reviewers are not obliged to read appendices. Reviewing is double-blind. We welcome complete results, work in progress, position papers, and preliminary or negative findings alike. Preliminary and exploratory work is also encouraged.
The workshop emphasizes interaction between academia and industry, providing opportunities for discussion beyond traditional conference presentations.
Contributions we especially encourage
These are encouraged contribution types, not separate tracks:
- Datasets, benchmarks, simulators, and temporal environments. We treat these as primary research artifacts. Benchmark submissions should report contamination audits and deduplication logs where applicable.
- Negative results, replication studies, and critical analyses that challenge prevailing assumptions about temporal foundation models.
- Deployment and operational experience reports from production forecasting and simulation systems, including failure modes absent from benchmark evaluation.
Scope
Topics
We welcome submissions across four connected research axes. Topics of interest include, but are not limited to, the directions below.
01
Forecasting and simulation tasks
Long-horizon and multi-resolution forecasting; multimodal contextual forecasting conditioned on text, video, events, covariates, or actions; calibrated probabilistic forecasting; prediction under sparse observations, regime changes, and distribution shift; scalable trajectory simulation; and adaptive, retrieval-augmented, or agentic forecasting systems, including model selection, ensembling, tool use, planning, and inference-time orchestration.
02
Temporal data and environments
Large-scale temporal pretraining corpora; environments combining time series with text, video, graphs, sensors, trajectories, actions, and events; video-as-environment corpora; physical evaluation suites; synthetic temporal data; simulation environments; benchmark realism; data quality; and scalable evaluation.
03
Temporal models
Time-aware models for irregular sampling; event-based models for marked point processes and transaction streams; hierarchical multi-timescale and state-space architectures; generative temporal models; time-series foundation models; scalable pretraining; multimodal fusion; memory and persistent state; embodied vision-language-action systems; and adaptation across domains, modalities, and horizons.
04
Evaluation and reliability
Benchmark realism, data quality, and leakage-aware evaluation; robustness under distribution shift and test-time adaptation; calibration and conformal uncertainty quantification; long-horizon and simulation consistency; neural scaling laws; reproducibility standards; and contamination audits for temporal datasets and pretraining corpora.
Evaluation
Review process
Submissions will be managed through OpenReview under standard conflict-of-interest policies. Each paper receives three reviews. We have confirmed 40+ reviewers across academia and industry; the program committee is listed on the Organizers page.
Poster and spotlight selections will balance technical quality, topical diversity, and interaction across the workshop communities.
Technical quality
Soundness of the method, analysis, evidence, and claims.
Evaluation rigor
Baselines, contamination and leakage awareness, and statistical treatment.
Reproducibility
Experimental transparency, artifact quality, and documentation.
Relevance and impact
Engagement with the four axes, including negative and replication results.
We will prioritise junior and early-career contributions in spotlight selection.
Participation
Policies
Non-archival
Per the NeurIPS 2026 workshop guidance, all workshop papers are non-archival and do not appear in proceedings. Accepted papers will be posted on OpenReview and listed on this site. Presenting here does not preclude later publication at an archival venue.
Concurrent and prior submissions
- Concurrent submission is permitted. The NeurIPS 2026 main-track handbook permits dual submissions to non-archival workshops.
- Previously published work is not eligible. Work already published at an ML or related conference, or accepted to the NeurIPS 2026 main conference, should not appear in the workshop.
- Preprints are fine. Posting to arXiv or a similar server does not affect eligibility, but the submitted PDF must remain anonymized.
Conflicts of interest
Organizers, and anyone with a personal conflict of interest with an organizer, may not submit to this workshop. Program committee members may submit; conflicted organizers are excluded from all discussion and decision-making for those papers.
Use of LLMs and AI assistants
We follow the policy in the NeurIPS 2026 main-track handbook:
- LLMs and agents cannot be authors.
- Important, original, or non-standard use of an LLM or agent must be documented. Routine spelling, grammar, editing, and basic code assistance do not require disclosure.
- Authors remain responsible for all text, figures, claims, and references.
- Prompt injection and attempts to manipulate reviewing are prohibited.
- Reviewers may not use LLMs. The NeurIPS workshop chairs opted out of the main-track reviewer-LLM experiment.
Submissions are also subject to the NeurIPS Code of Conduct and NeurIPS Ethics Guidelines.