Important Dates
| Submission | openreview |
| Paper Submission Deadline | Monday, February 3, 2025 (23:59 AoE) |
| Author notification | Wednesday, March 5, 2025 (23:59 AoE) |
| Camera-Ready Deadline | Wednesday, March 26, 2025 (23:59 AoE) |
| Workshop Day | Wednesday, April 27 or 28, 2025 |
Call for Papers
We are interested in contributions that push the boundaries of weight space learning, including both theoretical advancements and novel applications across various domains. We invite contributions from researchers and practitioners that explore the promise of weight space learning. Topics of interest include, but are not limited to:
Weight Space as a Modality
- Characterization of weight space properties such as symmetries (e.g., permutations, scaling, and beyond).
- Weight space augmentations, scaling laws, model zoo datasets, etc.
Weight Space Learning Tasks/Learning Paradigms
- Supervised approaches: Weight embeddings, meta-learning networks, (graph) hyper-networks.
- Unsupervised approaches: Autoencoders or hyper-representations.
- Weight space learning backbones: Plain MLPs, transformers, equivariant architectures (e.g., GNNs and neural functionals).
Theoretical Foundations
- Expressivity of weight space processing modules.
- Theoretical analysis of model weight properties.
- Generalization bounds of weight space learning methods.
Model/Weight Analysis
- Inferring model properties and behaviors from their weights.
- Investigating neural lineage and model trees through weights.
- Learning dynamics in population-based training.
- Interpretability of models via their weights.
Model/Weight Synthesis and Generation
- Modeling weight distributions to facilitate weight sampling.
- Generating weights in the context of transfer learning, learnable optimizers, implicit neural representation (INR) synthesis.
- Model operations/editing (e.g., model merging, model soups, model pruning, task arithmetic).
- Meta-learning and continual learning using model weights.
Applications of Weight Space Learning
- Computer vision tasks: Using NeRFs/INRs.
- Applications to physics and dynamical system modeling.
- Backdoor detection and adversarial robustness in weight space.
Tracks
We invite submissions in the form of extended abstracts (max 4 pages) or full papers (max 8 pages). The extended abstract track targets early-stage results with insightful findings that may foster discussion in the community. The full papers present substantiated results that advance the field of weight space learning. Accepted contributions will be presented in poster sessions and spotlight talks.
This year, ICLR is discontinuing the separate “Tiny Papers” track, and is instead requiring each workshop to accept short (3–5 pages in ICLR format, exact page length to be determined by each workshop) paper submissions, with an eye towards inclusion; see https://iclr.cc/Conferences/2025/CallForTinyPapers for more details. Authors of these papers will be earmarked for potential funding from ICLR, but need to submit a separate application for Financial Assistance that evaluates their eligibility. This application for Financial Assistance to attend ICLR 2025 will become available on https://iclr.cc/Conferences/2025/ at the beginning of February and close on March 2nd.
Submission:
Submissions are exclusively accepted via the official workshop submission portal on openreview.
Submission instructions
Submissions should follow the ICLR formatting guidelines, see the original ICLR call for papers and the workshop’s latex template. As a note, the reviewers will not be required to review the supplementary materials so make sure that your paper is self-contained. For the extended abstracts please use the same template but limit the submission to 4 pages, exclusive of references and supplementary materials. There will be an option on the submission site to differentiate full papers and extended abstracts.