Call for abstracts

FoPt 2025 welcomes submissions in a single non-archival track

The First Workshop on the Foundations of Post-training (FoPt 2025) invites submissions of abstracts addressing theoretical and empirical challenges in post-training, in a single non-archival track. Submitted abstracts can be up to one page long, and accepted abstracts will be presented as posters at the workshop. Paper submissions are due on Monday, May 19th, 2025.

The workshop solicits contributions addressing fundamental questions related to post-training in various levels of abstraction, including but not limited to the following topics of interest:

  1. Computational and statistical foundations of alignment
    • Fine-tuning algorithms
    • Inference-time methods
    • Interactions between pre-training and post-training
  2. Optimization landscape of post-training
    • Stability of fine-tuning methods
    • Convergence guarantees
  3. New and old models of supervision
    • Alternative forms of feedback, such as comparisons and corrections
    • Offline vs interactive feedback
    • Demonstration or chain-of-thought design
  4. Guarantees for emerging learning paradigms
    • Weak-to-strong generalization
    • Self-improvement and self-correction
    • Multi-agent learning
    • prompt or retrieval-based adaptation
  5. Novel frameworks and abstractions for post-training
    • Models for LLM reasoning abilities, such as chain-of-thought
    • Formalizations for new forms of generalization, e.g., in-context learning and length generalization
  6. Other topics
    • RL / bandits frameworks for LLM fine-tuning
    • Representational abilities of transformers
    • Theoretical foundations of LLMs

Important dates

(All dates are in 2025)

  • Submission deadline: Monday, May 19th, 5:00 PM EST
  • Author notification: Monday, May 26th
  • Workshop date: Monday, June 30th

Dual submission policy

Since the workshop is non-archival, FoPt invites submissions that may be substantially similar to abstracts/papers that have previously been published, accepted for publication, or submitted in parallel to a peer-reviewed conference or journal. Submitted abstracts need not be anonymized.

Submission Instructions

Formatting and anonymization

Submitted abstracts are limited to one page, excluding references. Authors are permitted to submit papers which have already previously appeared in a conference, journal or workshop, or published to Arxiv. There is no formatting constraint for such submissions.

Anonymization: Submissions should be suitable for single-blind reviewing; in particular, submissions are permitted to include author names and other identifying information. Reviewer names are withheld from the authors.

Style files: For submitted one-page abstracts, please use the COLT style file, linked below.

For papers which have previously appeared at a conference/journal/workshop/Arxiv, there is no page limit or formatting constraint.

Submitting your paper

Abstarcts should be submitted through OpenReview; the deadline for submissions is 5pm EST on Monday, May 19th, 2025.

OpenReview submission site: OpenReview