mlsanityData to 14 Oct 2026

Towards Global Neural Network Abstractions with Locally-Exact Reconstruction

arXiv 2210.12054 ↗First posted 21 Oct 2022cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

Neural networks are a powerful class of non-linear functions. However, their black-box nature makes it difficult to explain their behaviour and certify their safety. Abstraction techniques address this challenge by transforming the neural network into a simpler, over-approximated function. Unfortunately, existing abstraction techniques are slack, which limits their applicability to small local regions of the input domain. In this paper, we propose Global Interval Neural Network Abstractions with Center-Exact Reconstruction (GINNACER). Our novel abstraction technique produces sound over-approximation bounds over the whole input domain while guaranteeing exact reconstructions for any given local input. Our experiments show that GINNACER is several orders of magnitude tighter than state-of-the-art global abstraction techniques, while being competitive with local ones.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 37 of its 60 references that are in the map. First posted on arXiv 21 Oct 2022.

Builds on

37 of 60 references.

References · 37

  1. 2015Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Codingcited by 2,329
  2. 2017Reluplex: An Efficient SMT Solver for Verifying Deep Neural Networkscited by 744
  3. 2016Error bounds for approximations with deep ReLU networkscited by 574
  4. 2018Efficient Neural Network Robustness Certification with General Activation Functionscited by 445
  5. 2016Quantized Neural Networks: Training Neural Networks with Low Precision Weights and Activationscited by 426
  6. 2020Knowledge Distillation and Student-Teacher Learning for Visual Intelligence: A Review and New Outlookscited by 316
  7. 2018Model compression via distillation and quantizationcited by 289
  8. 2021Pruning and Quantization for Deep Neural Network Acceleration: A Surveycited by 263
  9. 2018Towards Fast Computation of Certified Robustness for ReLU Networkscited by 255
  10. 2019Algorithms for Verifying Deep Neural Networkscited by 221
  11. 2018Stochastic Activation Pruning for Robust Adversarial Defensecited by 171
  12. 2019A Convex Relaxation Barrier to Tight Robustness Verification of Neural Networkscited by 125
25 more
  1. 2021MLPerf Tiny Benchmarkcited by 124
  2. 2019Adversarial Robustness vs Model Compression, or Both?cited by 98
  3. 2019Defensive Quantization: When Efficiency Meets Robustnesscited by 95
  4. 2015Representation Benefits of Deep Feedforward Networkscited by 86
  5. 2018ResNet with one-neuron hidden layers is a Universal Approximatorcited by 79
  6. 2021PRIMA: General and Precise Neural Network Certification via Scalable Convex Hull Approximationscited by 77
  7. 2021The Second International Verification of Neural Networks Competition (VNN-COMP 2021): Summary and Resultscited by 71
  8. 2019An Abstraction-Based Framework for Neural Network Verificationcited by 68
  9. 2015Neural Network with Unbounded Activation Functions is Universal Approximatorcited by 63
  10. 2021Privacy-Preserving Machine Learning with Fully Homomorphic Encryption for Deep Neural Networkcited by 57
  11. 2018Mad Max: Affine Spline Insights into Deep Learningcited by 55
  12. 2021POLAR: A Polynomial Arithmetic Framework for Verifying Neural-Network Controlled Systemscited by 39
  13. 2020Lossless Compression of Deep Neural Networkscited by 34
  14. 2020Abstraction based Output Range Analysis for Neural Networkscited by 34
  15. 2019ReachNN: Reachability Analysis of Neural-Network Controlled Systemscited by 26
  16. 2021On the Effect of Pruning on Adversarial Robustnesscited by 23
  17. 2021Scaling Up Exact Neural Network Compression by ReLU Stabilitycited by 19
  18. 2019Sparse Polynomial Zonotopes: A Novel Set Representation for Reachability Analysiscited by 14
  19. 2020Interval Universal Approximation for Neural Networkscited by 14
  20. 2019Equivalent and Approximate Transformations of Deep Neural Networkscited by 11
  21. 2020Lossless Compression of Structured Convolutional Models via Liftingcited by 11
  22. 2022LinSyn: Synthesizing Tight Linear Bounds for Arbitrary Neural Network Activation Functionscited by 10
  23. 2019Neural network integral representations with the ReLU activation functioncited by 8
  24. 2022Formal Control Synthesis for Stochastic Neural Network Dynamic Modelscited by 7
  25. 2022Stability of Non-linear Neural Feedback Loops using Sum of Squarescited by 6

Built on by

Not cited yet.