Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

Provable Repair of Vision Transformers

Conference · · Lecture Notes in Computer Science AI Verification

Vision Transformers have emerged as state-of-the-art image recognition tools, but may still exhibit incorrect behavior. Incorrect image recognition can have disastrous consequences in safety-critical real-world applications such as self-driving automobiles. In this paper, we present Provable Repair of Vision Transformers (PRoViT), a provable repair approach that guarantees the correct classification of images in a repair set for a given Vision Transformer without modifying its architecture. PRoViT avoids negatively affecting correctly classified images (drawdown) by minimizing the changes made to the Vision Transformer’s parameters and original output. We observe that for Vision Transformers, unlike for other architectures such as ResNet or VGG, editing just the parameters in the last layer achieves correctness guarantees and very low drawdown. We introduce a novel method for editing these last-layer parameters that enables PRoViT to efficiently repair state-of-the-art Vision Transformers for thousands of ima

Research Organization:
University of California, Davis
Sponsoring Organization:
USDOE Office of Science (SC), Advanced Scientific Computing Research (ASCR)
DOE Contract Number:
SC0022285
OSTI ID:
2476499
Journal Information:
Lecture Notes in Computer Science AI Verification, Journal Name: Lecture Notes in Computer Science AI Verification; ISSN 0302-9743
Country of Publication:
United States
Language:
English

References (7)

Architecture-Preserving Provable Repair of Deep Neural Networks journal June 2023
Identifying Medical Diagnoses and Treatable Diseases by Image-Based Deep Learning journal February 2018
Fast and precise certification of transformers
  • Bonaert, Gregory; Dimitrov, Dimitar I.; Baader, Maximilian
  • Proceedings of the 42nd ACM SIGPLAN International Conference on Programming Language Design and Implementation https://doi.org/10.1145/3453483.3454056
conference June 2021
An abstract domain for certifying neural networks journal January 2019
Provable repair of deep neural networks conference June 2021
Deep Residual Learning for Image Recognition conference June 2016
Measuring Catastrophic Forgetting in Neural Networks journal April 2018

Similar Records

Provable Repair of Vision Transformers
Journal Article · 2024 · Lecture Notes in Computer Science · OSTI ID:2516699

Architecture-Preserving Provable Repair of Deep Neural Networks
Journal Article · 2023 · Proceedings of the ACM on Programming Languages · OSTI ID:1975934

Topical meeting on machine vision (summaries)
Conference · 1986 · OSTI ID:6112293