mlsanityData to 7 Oct 2026

Predicting Neural Network Accuracy from Weights

arXiv 2002.11448 ↗First posted 26 Feb 2020stat.ML cs.LGcited by 107

Not on the map. Why?

Abstract (arXiv)

We show experimentally that the accuracy of a trained neural network can be predicted surprisingly well by looking only at its weights, without evaluating it on input data. We motivate this task and introduce a formal setting for it. Even when using simple statistics of the weights, the predictors are able to rank neural networks by their performance with very high accuracy (R2 score more than 0.98). Furthermore, the predictors are able to rank networks trained on different, unobserved datasets and with different architectures. We release a collection of 120k convolutional neural networks trained on four different datasets to encourage further research in this area, with the goal of understanding network training and performance better.

Key facts. Cited by 107 papers in the map (data through 7 Oct 2026). First posted on arXiv 26 Feb 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 11; 2022: 11; 2023: 18; 2024: 22; 2025: 24; 2026: 21.

107 papers.

Sort:

By area:

  1. 2022From data to functa: Your data point is a function and you can treat it like onecited by 183
  2. 2021Deep Learning Through the Lens of Example Difficultycited by 139
  3. 2021Assessing Generalization of SGD via Disagreementcited by 116
  4. 2023Equivariant Architectures for Learning in Deep Weight Spacescited by 106
  5. 2023Permutation Equivariant Neural Functionalscited by 69
  6. 2021What Does Rotation Prediction Tell Us about Classifier Accuracy under Varying Testing Environments?cited by 63
  7. 2022Hyper-Representations as Generative Models: Sampling Unseen Neural Network Weightscited by 63
  8. 2023Deep Learning on Implicit Neural Representations of Shapescited by 62
  9. 2021A Theory of PAC Learnability of Partial Concept Classescited by 59
  10. 2024Graph Neural Networks for Learning Equivariant Representations of Neural Networkscited by 59
  11. 2023Graph Metanetworks for Processing Diverse Neural Architecturescited by 55
  12. 2023Neural Functional Transformerscited by 50
38 more
  1. 2021Scalable Diverse Model Selection for Accessible Transfer Learningcited by 40
  2. 2024Towards Scalable and Versatile Weight Space Learningcited by 40
  3. 2022Model Zoos: A Dataset of Diverse Populations of Neural Network Modelscited by 38
  4. 2023Equivariant Deep Weight Space Alignmentcited by 34
  5. 2025Symmetry in Neural Network Parameter Spacescited by 31
  6. 2024Universal Neural Functionalscited by 25
  7. 2025We Should Chart an Atlas of All the World's Modelscited by 25
  8. 2024Improved Generalization of Weight Space Networks via Augmentationscited by 24
  9. 2022On the Strong Correlation Between Model Invariance and Generalizationcited by 23
  10. 2024Scale Equivariant Graph Metanetworkscited by 23
  11. 2023Confidence and Dispersity Speak: Characterising Prediction Matrix for Unsupervised Accuracy Estimationcited by 22
  12. 2026SHINE: A Scalable In-Context Hypernetwork for Mapping Context to LoRA in a Single Passcited by 22
  13. 2022Multi-Object Navigation with dynamically learned neural implicit representationscited by 21
  14. 2024Position: Leverage Foundational Models for Black-Box Optimizationcited by 21
  15. 2024The Empirical Impact of Neural Parameter Symmetries, or Lack Thereofcited by 21
  16. 2024Learning on LoRAs: GL-Equivariant Processing of Low-Rank Weight Spaces for Large Finetuned Modelscited by 21
  17. 2023Privacy and Efficiency of Communications in Federated Split Learningcited by 20
  18. 2024Deep Linear Probe Generators for Weight Space Learningcited by 17
  19. 2024Energy-based Automated Model Evaluationcited by 16
  20. 2023On the Importance of Feature Separability in Predicting Out-Of-Distribution Errorcited by 15
  21. 2024Monomial Matrix Group Equivariant Neural Functional Networkscited by 15
  22. 2022On the Relationship Between Explanation and Prediction: A Causal Viewcited by 14
  23. 2024Learning Useful Representations of Recurrent Neural Network Weight Matricescited by 13
  24. 2022Hyper-Representations for Pre-Training and Transfer Learningcited by 12
  25. 2023LOVM: Language-Only Vision Model Selectioncited by 12
  26. 2023How to Train Neural Field Representations: A Comprehensive Study and Benchmarkcited by 12
  27. 2024Unsupervised Model Tree Heritage Recoverycited by 11
  28. 2021Hyper-Representations: Self-Supervised Representation Learning on Neural Network Weights for Model Characteristic Predictioncited by 9
  29. 2023Set-based Neural Network Encoding Without Weight Tyingcited by 9
  30. 2025Predicting the Performance of Black-box LLMs through Follow-up Queriescited by 9
  31. 2021Understanding and Accelerating Neural Architecture Search with Training-Free and Theory-Grounded Metricscited by 7
  32. 2024Symmetry Discovery for Different Data Typescited by 7
  33. 2024Learning on Model Weights using Tree Expertscited by 7
  34. 2025GradMetaNet: An Equivariant Architecture for Learning on Gradientscited by 7
  35. 2023CAME: Contrastive Automated Model Evaluationcited by 6
  36. 2024Connecting NeRFs, Images, and Textcited by 6
  37. 2024LLaNA: Large Language and NeRF Assistantcited by 6
  38. 2023Deep Learning on Object-centric 3D Neural Fieldscited by 5

50 most cited of 107 All 107 (JSON)