Important Dates
| Submission portal | OpenReview |
| Template | Download ZIP |
| Paper submission deadline | |
| Author notification | September 29, 2026 (Anywhere on Earth) |
| Camera-ready deadline | To be announced |
| Workshop day | To be announced |
Call for Papers
We invite contributions that treat neural network artifacts—including weights, gradients, optimization trajectories, internal representations, and other computational traces—as a learning data modality. We welcome theoretical, methodological, empirical, and applied work. Topics include, but are not limited to:
Datasets and Benchmarks
- Standardized model zoos and neural-artifact datasets.
- Evaluation protocols and new benchmark tasks.
- Tools and infrastructure for collecting, documenting, and sharing model populations.
Foundations and Theory
- Structure, symmetries, invariances, and scaling laws of neural artifacts.
- Theoretical frameworks for learning on weights and computational traces.
- Specialized architectures, including equivariant metanetworks and metamodels.
- Expressivity, generalization, and optimization in weight space.
Model Analysis and Dynamics
- Predicting performance, generalization, safety, robustness, fairness, memorization, or backdoors from artifacts.
- Understanding optimization trajectories and learning dynamics.
- Interpretability through weights, activations, representations, gradients, and other traces.
- Neural lineage, provenance, and relationships among models.
Model Synthesis and Control
- Hypernetworks, parameter generation, and learned optimizers.
- Model merging, task arithmetic, model editing, pruning, and steering.
- Neural field and implicit neural representation synthesis.
- Model safety and reliability interventions.
Model Search and Selection
- Navigating model populations for inference, fine-tuning, and transfer learning.
- Predicting compatibility or interference between models.
- Efficient model selection without expensive retraining or evaluation.
Model Populations and AI Supply Chains
- Mapping and visualizing model ecosystems and model atlases.
- Studying model lineages, emerging trends, and knowledge gaps.
- Analyzing AI supply chains and their effects on model populations and weights.
- Population-level studies spanning research communities and deployment contexts.
Tracks
We will offer two submission tracks:
- Extended abstracts (4–6 pages): Early-stage results, position papers, new ideas, negative results, benchmark proposals, and other contributions that can foster discussion in the community.
- Full papers (8–12 pages): Substantiated research contributions that advance the study of neural artifacts and weight-space learning.
Page limits exclude references and supplementary material. Accepted contributions will be considered for poster presentations and spotlight talks. Further details will be announced when submissions open.
Submission
Submissions will be accepted through the OpenReview submission portal.
All submissions are non-archival. Authors may choose to make their submission public on OpenReview.
Submission instructions
Submissions should follow the NeurIPS 2026 formatting and generative-AI guidelines. Papers must be self-contained; reviewers will not be required to consult supplementary material. The official template, policies, anonymity requirements, and track-selection instructions will be linked here when available.