Call for posters
We invite abstracts for poster presentation on any topic within the scope below. The workshop is non-archival: work in progress, negative results, and work already published or under review elsewhere are all welcome. Abstracts are reviewed for fit with the workshop only. Students and early-career researchers are particularly encouraged to submit.
Scope
In scope is any work that studies what happens to optimization when the trainable architecture or sparse topology changes during training. Examples include:
- Growing neurons, layers or whole blocks during training
- Stacking and progressive scaling of transformers
- Dynamic sparse training and prune-and-regrow methods
- Modular or capacity expansion in continual and reinforcement learning
- Transferring parameters, optimizer state and normalization statistics across structural changes
- Deciding when and where to grow, and measuring the quality of a growth transition
Out of scope:
- Classical neural architecture search with independently trained candidates (unless the focus is transferring training state between candidates, as in DARTS)
- Input-dependent conditional computation, such as mixture-of-experts routing, without any structural change during training
Format
- Submission: an abstract of up to 300 words. For previously published work, you may instead submit the paper itself as a PDF (no page limit).
- Poster: A0, portrait orientation.
- Posters are presented during the lunch and poster session.
Important dates
| Poster abstract deadline | 27 November 2026 |
|---|---|
| Notification of acceptance | 30 November 2026 |
| Registration closes | 2 December 2026 |
| Workshop | 7 December 2026 |