Call for papers

CFP: Latent & Implicit Thinking – Going Beyond CoT Reasoning (LIT) Workshop at ICLR 2026

We invite original research paper submissions on latent and implicit reasoning in vision and language models. The workshop seeks to unify diverse yet complementary research directions exploring alternative forms of intermediate reasoning beyond explicit Chain-of-Thought (CoT) reasoning.

Topics of Interest

We welcome submissions on, but not limited to, the following themes:

Training Strategy for Implicit Reasoning

e.g., curriculum learning, distillation, reinforcement learning, and pretraining from scratch.

Looped Architectures

Recurrence mechanisms (loop unrolling, dynamic halting). Training curricula and stability for deep iterative models.

Mechanical Interpretability of Implicit Reasoning

Layer-wise specialization and attribution of reasoning functions. Causal interventions on intermediate representations.

Special Thinking Tokens

Text CoT compressed to special tokens (e.g., continuous thought tokens, VQ-VAE codes, gist tokens). CoT augmentation via filler or planning tokens.

KV-Cache and Hybrid Stateful Reasoning

Leveraging key/value caches for multi-step latent inference. Comparisons between pure activation vs. cache-augmented loops.

Inference Paradigms Beyond Autoregression

Text diffusion for bidirectional, iterative denoising-based reasoning. Fractal generative frameworks and next-block prediction.

Theoretical Results on Depth, Scaling, and Efficiency

Theoretical bounds on reasoning depth vs. layer count. Parameter- and compute-efficient designs for implicit reasoning.

Evaluation and Benchmarks

Metrics and probes for implicit vs. explicit CoT capabilities. Datasets and tasks that stress ultra-deep or multi-hop latent reasoning.

Limitations and Safety

Understanding pros and cons of implicit and explicit CoT, and interpretability & faithfulness of reasoning from a safety and alignment perspective.

Submission Guidelines

The workshop will accept papers from preliminary research results and visionary papers to full-length papers. We will host a Tiny and Short Paper Track in alignment with ICLR’s initiative, explicitly welcoming late-breaking results, replication studies, and conceptual explorations that may not yet have full-paper maturity. Review criteria will emphasize clarity, originality, and potential impact rather than extensive experiments.

Submissions should follow the ICLR proceedings format and choose the suitable categories as follows:

  • Tiny and Short Paper: 2 to 4 pages + references and appendix.
  • Regular Paper: up to 9 pages + references and appendix.

For novelty, the workshop does not accept submissions that have previously been published at ICLR or other machine learning or related venues. For openness, we encourage submissions with sufficient open-source resources (e.g., checkpoint, code, data, training details). We do not allow AI as primary authors for workshop submissions.

All accepted papers will be presented as posters. We will select around 3 papers for short oral presentations and 2 papers for outstanding paper awards with potential cash incentives. To support accessibility, we will publish accepted works on the workshop website and highlight selected tiny papers during the poster session, ensuring visibility for early-stage contributors. While the workshop is non-archival, the workshop webpage will be maintained to hold all related materials of the workshop for future online viewers.

Important Dates

  • Workshop website, schedule, and call for papers release: December 2025
  • Submission deadline: January 30, 2026
  • Reviewing period: January 30 - February 28, 2026
  • Notification of acceptance: March 1, 2026, 11:59pm AoE
  • Camera-ready deadline: March 10, 2026
  • Workshop date: April/May 2026 (TBA - during ICLR 2026 main conference week)

Submission Site

Submissions will be accepted through OpenReview. The submission link will be available soon.

Note: The OpenReview venue will be created by end of December 2025. Once available, the submission link will be updated here.

LLM Usage Policy

In alignment with the ICLR 2026 Policies on Large Language Model Usage, our workshop will clearly distinguish between AI-assisted and AI-generated content. We will explicitly prohibit AI-generated submissions in the tiny and short paper tracks, while allowing limited, transparent AI assistance (e.g., for grammar correction or rephrasing) as long as the intellectual contribution and analysis remain primarily human-authored. During the review process, LLMs will not be allowed for automated reviewing or decision-making; all evaluations will be conducted by human reviewers. The organizers may use LLMs only for logistical purposes, such as drafting website descriptions or formatting templates, with full human oversight. Any substantial use of AI tools by authors, reviewers, or organizers must be disclosed in submission metadata or acknowledgments.

FAQ

Will the OpenReview be open to the public?

We will implement the following policy:

  1. All submissions will be private – to their respective authors, to the reviewers assigned to review them, and to the workshop organizers – until the camera-ready due date.
  2. We will communicate the accept/reject decisions individually and privately to the respective authors.
  3. Accepted submissions (abstract + pdf) will be publicly visible after the notification.
  4. Rejected and withdrawn submissions will not become public.
  5. Reviews will not be made public. They will be visible only to the authors, the reviewers who will have written them, and the organizers.
Anonymity requirements

Submissions should be anonymized for a double-blind review. We do not set an anonymity period.

That means you must submit your paper pdf with author names and affiliations removed, but you can still share it publicly on platforms like ArXiv or on social media.

Dual submission

We accept submissions of the following types:

  1. Fully original work not under review elsewhere.
  2. Work that is under review at another venue (e.g., ICML, ARR, including ICLR 2026). If you submit, make sure you are not violating the submission guidelines and anonymity requirements of the other venue(s).
  3. Works accepted at other venues, we still require submitting the paper through the portal and going through the normal reviewing process, to decide if it will be accepted and presented at LIT.

For any other questions, please contact the organizers.

Are we allowed to submit a paper with an appendix?

Yes, you can submit an appendix but reviewers are not required to review it.

Student Funding

We plan to offer a limited number of travel grants to support attendees from underrepresented groups, including but not limited to individuals with disabilities, those from developing countries, and early-career researchers facing financial constraints.

Awards and prizes

We will select around 3 papers for short oral presentations and 2 papers for outstanding paper awards with potential cash incentives. The details will be announced later.