Extended abstracts
Early-stage results, position papers, new ideas, negative results, benchmark proposals, and other contributions that can foster discussion in the community; 4–6 pages (excluding references and supplementary material).
Machine Learning/CORE Unranked
NeuralArtifacts 2026
Neural Network Artifacts as a New Data Modality
The NeurIPS 2026 Workshop on Neural Network Artifacts as a New Data Modality explores neural network artifacts—such as weights, gradients, and optimization trajectories—as a novel data modality for machine learning. It aims to foster research on learning from populations of models to enable tasks like model search, synthesis, analysis, and understanding AI supply chains, building on the inaugural ICLR 2025 workshop.
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Full papers
Substantiated research contributions that advance the study of neural artifacts and weight-space learning; 8–12 pages (excluding references and supplementary material).
Research areas in scope
01
Datasets and Benchmarks
Standardized model zoos and neural-artifact datasetsEvaluation protocols and new benchmark tasksTools and infrastructure for collecting, documenting, and sharing model populations
02
Foundations and Theory
Structure, symmetries, invariances, and scaling laws of neural artifactsTheoretical frameworks for learning on weights and computational tracesSpecialized architectures, including equivariant metanetworks and metamodelsExpressivity, generalization, and optimization in weight space
03
Model Analysis and Dynamics
Predicting performance, generalization, safety, robustness, fairness, memorization, or backdoors from artifactsUnderstanding optimization trajectories and learning dynamicsInterpretability through weights, activations, representations, gradients, and other tracesNeural lineage, provenance, and relationships among models
04
Model Synthesis and Control
Hypernetworks, parameter generation, and learned optimizersModel merging, task arithmetic, model editing, pruning, and steeringNeural field and implicit neural representation synthesisModel safety and reliability interventions
05
Model Search and Selection
Navigating model populations for inference, fine-tuning, and transfer learningPredicting compatibility or interference between modelsEfficient model selection without expensive retraining or evaluation
06
Model Populations and AI Supply Chains
Mapping and visualizing model ecosystems and model atlasesStudying model lineages, emerging trends, and knowledge gapsAnalyzing AI supply chains and their effects on model populations and weightsPopulation-level studies spanning research communities and deployment contexts
Policies worth checking twice
- Submissions are non-archival.
- Authors may choose to make their submission public on OpenReview.
- Submissions must follow the NeurIPS 2026 formatting and generative-AI guidelines.
- Papers must be self-contained; reviewers will not be required to consult supplementary material.
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026