Fast and precise certification of transformers

Bonaert, Gregory; Dimitrov, Dimitar; Baader, Maximilian; Vechev, Martin

doi:10.1145/3453483.3454056

Cited by 11 publications

(8 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…CROWN-BaF (Shi et al 2020) addresses the above issue by using the forward mode to handle non-linear functions in attention layers. Unlike CROWN-BaF, DeepT (Bonaert et al 2021), which is based on abstract interpretation and Zonotopes, improves efficiency via noise symbol reduction. Prior work shows that DeepT is more precise than CROWN-BaF but consumes more time.…”

Section: Contributions Our Main Contributions Arementioning

confidence: 99%

See 1 more Smart Citation

GaLileo: General Linear Relaxation Framework for Tightening Robustness Certification of Transformers

Zhang,

Shen,

Guo

et al. 2024

AAAI

View full text Add to dashboard Cite

Transformers based on attention mechanisms exhibit vulnerability to adversarial examples, posing a substantial threat to the security of their applications. Aiming to solve this problem, the concept of robustness certification is introduced to formally ascertain the presence of any adversarial example within a specified region surrounding a given sample. However, prior works have neglected the dependencies among inputs of softmax (the most complex function in attention mechanisms) during linear relaxations. This oversight has consequently led to imprecise certification results. In this work, we introduce GaLileo, a general linear relaxation framework designed to certify the robustness of Transformers. GaLileo effectively surmounts the trade-off between precision and efficiency in robustness certification through our innovative n-dimensional relaxation approach. Notably, our relaxation technique represents a pioneering effort as the first linear relaxation for n-dimensional functions such as softmax. Our novel approach successfully transcends the challenges posed by the curse of dimensionality inherent in linear relaxations, thereby enhancing linear bounds by incorporating input dependencies. Our evaluations encompassed a thorough analysis utilizing the SST and Yelp datasets along with diverse Transformers of different depths and widths. The experimental results demonstrate that, as compared to the baseline method CROWN-BaF, GaLileo achieves up to 3.24 times larger certified radii while requiring similar running times. Additionally, GaLileo successfully attains certification for Transformers' robustness against multi-word lp perturbations, marking a notable accomplishment in this field.

show abstract

Section: Contributions Our Main Contributions Arementioning

confidence: 99%

“…Limitation of prior works. Indeed, prior works (Shi et al 2020;Bonaert et al 2021) have made valuable contributions to the robustness certification of Transformers. However, we contend that their precision is compromised due to the loose relaxations applied to softmax functions.…”

Section: Introductionmentioning

confidence: 99%

GaLileo: General Linear Relaxation Framework for Tightening Robustness Certification of Transformers

Zhang,

Shen,

Guo

et al. 2024

AAAI

View full text Add to dashboard Cite

show abstract

“…SoftMax function [10][11][12] has a wide range of applications in machine learning and deep learning, mainly for multiclassification problems. Most of the time, in the multi-classification problem, we want to output the probability of getting a certain category.…”

Section: Anti-domain Adaptation Of Knowledge Mappingmentioning

confidence: 99%

Research on fault diagnosis model driven by artificial intelligence from domain adaptation to domain generalization

Xu,

Cao

2023

Fourth International Conference on Artificial Intelligence and Electromechanical Automation (AIEA 2023)

View full text Add to dashboard Cite

Transfer learning is studied through inductive research, knowledge transfer, lifelong learning and cumulative learning. The research scope and field of transfer learning are very wide. Transfer learning is a training method that can transfer the network structure and weight originally used to solve the mature task methodology to the new learning task, and can also get better results in the new task. The reason why transfer learning can be realized is that the characteristics of convolutional neural network in shallow learning are universal. In the case of insufficient samples, transfer learning can be used to transfer these general feature learning from other trained networks, so as to save training time and obtain better recognition results.

show abstract

“…DeepZ is a novel generic, point-wise zonotope abstract transformers for ReLU, Sigmoid, and Tanh activations which is more scalable and fast. Another work in this direction is proposed by [9] where authors proposed DeepT an abstract transformer-based network certification method. They attempted to certify larger transformers against synonym replacement based attacks.…”

Section: Robustness By Certificationmentioning

confidence: 99%

A survey in Adversarial Defences and Robustness in NLP

Goyal¹,

Doddapaneni²,

Khapra³

et al. 2022

Preprint

View full text Add to dashboard Cite

In recent years, it has been seen that deep neural networks are lacking robustness and are likely to break in case of adversarial perturbations in input data. Strong adversarial attacks are proposed by various authors for computer vision and Natural Language Processing (NLP). As a counter-effort, several defense mechanisms are also proposed to save these networks from failing. In contrast with image data, generating adversarial attacks and defending these models is not easy in NLP because of the discrete nature of the text data. However, numerous methods for adversarial defense are proposed of late, for different NLP tasks such as text classification, named entity recognition, natural language inferencing, etc. These methods are not just used for defending neural networks from adversarial attacks, but also used as a regularization mechanism during training, saving the model from overfitting. The proposed survey is an attempt to review different methods proposed for adversarial defenses in NLP in the recent past by proposing a novel taxonomy. This survey also highlights the fragility of the advanced deep neural networks in NLP and the challenges in defending them.

show abstract

Fast and precise certification of transformers

Cited by 11 publications

References 20 publications

GaLileo: General Linear Relaxation Framework for Tightening Robustness Certification of Transformers

GaLileo: General Linear Relaxation Framework for Tightening Robustness Certification of Transformers

Research on fault diagnosis model driven by artificial intelligence from domain adaptation to domain generalization

A survey in Adversarial Defences and Robustness in NLP

Contact Info

Product

Resources

About