mlsanityData to 7 Oct 2026

Towards White Box Deep Learning

arXiv 2403.09863 ↗First posted 14 Mar 2024cs.LG cs.AI cs.NEcited by 0

Not on the map. Why?

Abstract (arXiv)

Deep neural networks learn fragile "shortcut" features, rendering them difficult to interpret (black box) and vulnerable to adversarial attacks. This paper proposes semantic features as a general architectural solution to this problem. The main idea is to make features locality-sensitive in the adequate semantic topology of the domain, thus introducing a strong regularization. The proof of concept network is lightweight, inherently interpretable and achieves almost human-level adversarial test metrics - with no adversarial training! These results and the general nature of the approach warrant further research on semantic features. The code is available at https://github.com/314-Foundation/white-box-nn

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 32 of its 39 references that are in the map. First posted on arXiv 14 Mar 2024.

Builds on

32 of 39 references.

References · 32

  1. 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,326
  2. 2013Intriguing properties of neural networkscited by 5,114
  3. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 2,914
  4. 2020Shortcut Learning in Deep Neural Networkscited by 1,928
  5. 2020Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attackscited by 1,442
  6. 2018Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examplescited by 1,259
  7. 2018Sanity Checks for Saliency Mapscited by 1,149
  8. 2019Adversarial Examples Are Not Bugs, They Are Featurescited by 950
  9. 2018This Looks Like That: Deep Learning for Interpretable Image Recognitioncited by 731
  10. 2017Dynamic Routing Between Capsulescited by 689
  11. 2018Towards Robust Interpretability with Self-Explaining Neural Networkscited by 505
  12. 2020Overfitting in adversarially robust deep learningcited by 504
20 more
  1. 2017Decision-Based Adversarial Attacks: Reliable Attacks Against Black-Box Machine Learning Modelscited by 459
  2. 2017Interpretation of Neural Networks is Fragilecited by 443
  3. 2019On Evaluating Adversarial Robustnesscited by 442
  4. 2018Adversarially Robust Generalization Requires More Datacited by 318
  5. 2017Adversarial Examples for Semantic Segmentation and Object Detectioncited by 275
  6. 2018Audio Adversarial Examples: Targeted Attacks on Speech-to-Textcited by 268
  7. 2019Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNetcited by 260
  8. 2018Towards Fast Computation of Certified Robustness for ReLU Networkscited by 254
  9. 2017Deep Learning for Case-Based Reasoning through Prototypes: A Neural Network that Explains Its Predictionscited by 234
  10. 2019Kornia: an Open Source Differentiable Computer Vision Library for PyTorchcited by 231
  11. 2019Adversarial Policies: Attacking Deep Reinforcement Learningcited by 209
  12. 2018Towards the first adversarially robust neural network model on MNISTcited by 104
  13. 2022Understanding Robust Overfitting of Adversarial Training and Beyondcited by 50
  14. 2023Adversarial attacks and defenses in explainable artificial intelligence: A surveycited by 47
  15. 2021Adversarial Attacks for Tabular Data: Application to Fraud Detection and Imbalanced Datacited by 45
  16. 2019Are Perceptually-Aligned Gradients a General Property of Robust Classifiers?cited by 42
  17. 2021Analysis and Applications of Class-wise Robustness in Adversarial Trainingcited by 33
  18. 2022Why Robust Generalization in Deep Learning is Difficult: Perspective of Expressive Powercited by 23
  19. 2021Self-Interpretable Model with TransformationEquivariant Interpretationcited by 21
  20. 2023Which Models have Perceptually-Aligned Gradients? An Explanation via Off-Manifold Robustnesscited by 20

Built on by

Not cited yet.