Filters








5,506 Hits in 5.1 sec

Formal Security Analysis of Neural Networks using Symbolic Intervals [article]

Shiqi Wang, Kexin Pei, Justin Whitehouse, Junfeng Yang, Suman Jana
2018 arXiv   pre-print
Due to the increasing deployment of Deep Neural Networks (DNNs) in real-world security-critical domains including autonomous vehicles and collision avoidance systems, formally checking security properties  ...  Our experiments demonstrate that symbolic interval analysis is a promising new direction towards rigorously analyzing different security properties of DNNs.  ...  Any opinions, findings, conclusions, or recommendations expressed herein are those of the authors, and do not necessarily reflect those of the US Government, ONR, or NSF.  ... 
arXiv:1804.10829v3 fatcat:3t427gxoovdarbgbyf7u4kjjmy

Efficient Formal Safety Analysis of Neural Networks [article]

Shiqi Wang, Kexin Pei, Justin Whitehouse, Junfeng Yang, Suman Jana
2018 arXiv   pre-print
Thus, there is an urgent need for formal analysis systems that can rigorously check neural networks for violations of different safety properties such as robustness against adversarial perturbations within  ...  neural networks.  ...  Any opinions, findings, conclusions, or recommendations expressed herein are those of the authors, and do not necessarily reflect those of the US Government, ONR, or NSF.  ... 
arXiv:1809.08098v3 fatcat:j6kjfrohqnecrkmqyzg2lqrdja

ReluDiff: Differential Verification of Deep Neural Networks [article]

Brandon Paulsen, Jingbo Wang, Chao Wang
2020 arXiv   pre-print
Our method consists of a fast but approximate forward interval analysis pass followed by a backward pass that iteratively refines the approximation until the desired property is verified.  ...  As deep neural networks are increasingly being deployed in practice, their efficiency has become an important issue.  ...  PRELIMINARIES First, we review the basics of interval analysis for neural networks.  ... 
arXiv:2001.03662v1 fatcat:xad4hdubxbgrvopozpdcf3jtt4

The illusion of structure or insufficiency of approach? the un(3) of unruly problems

Bradly Alicea, Jesse Parent, Ankit Gupta
2020 Figshare  
We call this class of problemshard-to-represent, and are characterized by the difficulties of quantification and symbolization, as well as the inherent un-physicality of a system.  ...  To counter these difficulties, we propose a new analytical paradigm called perceptual analysis, which brings an umbrella of diverse approaches to bear.  ...  Acknowledgments We would like to thank the Saturday Morning NeuroSim group (YouTube: https://tinyurl.com/Sat-Morning-NeuroSim ) for their comments and time to discuss the development of this essay.  ... 
doi:10.6084/m9.figshare.11998731 fatcat:aujgoms3xnd2nn7f4ebvqfmese

A Review of Formal Methods applied to Machine Learning [article]

Caterina Urban, Antoine Miné
2021 arXiv   pre-print
The large majority of them verify trained neural networks and employ either SMT, optimization, or abstract interpretation techniques.  ...  We review state-of-the-art formal methods applied to the emerging field of the verification of machine learning systems.  ...  The neural network is encoded as a Boolean combination of linear constraints using a Cartesian product of intervals.  ... 
arXiv:2104.02466v2 fatcat:6ghs5huoynbc5h7lndajmsoxyu

PEREGRiNN: Penalized-Relaxation Greedy Neural Network Verifier [article]

Haitham Khedr, James Ferlez, Yasser Shoukry
2021 arXiv   pre-print
In this paper, we introduce a new approach to formally verify the most commonly considered safety specifications for ReLU NNs -- i.e. polytopic specifications on the input and output of the network.  ...  As a consequence, techniques of formal verification have been recognized as crucial to the design and deployment of safe NNs.  ...  .: Formal security analysis of neural networks using symbolic intervals. In: 27th {USENIX} Security Symposium ({USENIX} Security 18). pp. 1599-1614 (2018) 32.  ... 
arXiv:2006.10864v2 fatcat:fpxkh5p62rg55jaym62haehgm4

Beyond Robustness: Resilience Verification of Tree-Based Classifiers [article]

Stefano Calzavara, Lorenzo Cazzaro, Claudio Lucchese, Federico Marcuzzi, Salvatore Orlando
2021 arXiv   pre-print
Our results show that resilience verification is useful and feasible in practice, yielding a more reliable security assessment of both standard and robust decision tree models.  ...  We then introduce a formally sound data-independent stability analysis for decision trees and decision tree ensembles, which we experimentally assess on public datasets and we leverage for resilience verification  ...  Preliminaries neural networks, which provide robustness guarantees for all Our analysis leverages intervals of real numbers.  ... 
arXiv:2112.02705v1 fatcat:ahw6lbkf7fbnnlubo7z5zuq4wy

MixTrain: Scalable Training of Verifiably Robust Neural Networks [article]

Shiqi Wang, Yizheng Chen, Ahmed Abdou, Suman Jana
2018 arXiv   pre-print
The adversarially robust training only makes the networks robust against a subclass of attackers and we reveal such weaknesses by developing a new attack based on interval gradients.  ...  Making neural networks robust against adversarial inputs has resulted in an arms race between new defenses and attacks.  ...  Sound over-approximation techniques like symbolic interval analysis provides the lower and upper bounds (in terms of two parallel linear equations over symbolic inputs) of a neural network's output for  ... 
arXiv:1811.02625v2 fatcat:yznavdhrvrbyzjfjd2xeh5aj6a

Neural and Bayesian Networks to Fight Crime: the NBNC Meta-Model of Risk Analysis [chapter]

Gaetano Bruno
2011 Artificial Neural Networks - Application  
Currently, security officers of the Italian banking system are using the three risk indexes in their analysis and robbery risk management.  ...  of the NBNC meta-model; • a brief discussion about the method used in order to derivate a Bayesian network from a database through an ANN; www.intechopen.com Artificial Neural Networks -Application  ...  Neural and Bayesian Networks to Fight Crime: the NBNC Meta-Model of Risk Analysis, Artificial Neural Networks -Application, Dr.  ... 
doi:10.5772/15444 fatcat:dycuich2wvaafpcasyu6bsnbn4

Combining Graph Neural Networks with Expert Knowledge for Smart Contract Vulnerability Detection

Zhenguang Liu, Peng Qian, Xiaoyang Wang, Yuan Zhuang, Lin Qiu, Xun Wang
2021 IEEE Transactions on Knowledge and Data Engineering  
Recent researches focus on the symbolic execution and formal analysis of smart contracts for vulnerability detection, yet to achieve a precise and scalable solution.  ...  In this paper, we explore using graph neural networks and expert knowledge for smart contract vulnerability detection.  ...  Graph Neural Network With the remarkable success of neural networks, graph neural network has been investigated extensively in various fields such as graph classification [32, 33] , program analysis  ... 
doi:10.1109/tkde.2021.3095196 fatcat:ta2jlqgyk5gyzaov5orldr3dnm

Scalable Verification of Quantized Neural Networks (Technical Report) [article]

Thomas A. Henzinger, Mathias Lechner, Đorđe Žikelić
2022 arXiv   pre-print
Formal verification of neural networks is an active topic of research, and recent advances have significantly increased the size of the networks that verification tools can handle.  ...  In particular, we propose three techniques for making SMT-based verification of quantized neural networks more scalable.  ...  Efficient formal safety analysis of neural networks . Wang, S.; Pei, K.; Whitehouse, J.; Yang, J.; and Jana, S. 2018b. Formal security analysis of neural networks using symbolic intervals.  ... 
arXiv:2012.08185v2 fatcat:wk6y2yzk3ncttaaeovqbu4fj74

DiffRNN: Differential Verification of Recurrent Neural Networks [article]

Sara Mohammadinejad, Brandon Paulsen, Chao Wang, Jyotirmoy V. Deshmukh
2020 arXiv   pre-print
As the memory footprint and energy consumption of such components become a bottleneck, there is interest in compressing and optimizing such networks using a range of heuristic techniques.  ...  Recurrent neural networks (RNNs) such as Long Short Term Memory (LSTM) networks have become popular in a variety of applications such as image processing, data classification, speech recognition, and as  ...  DIFFRNN leverages interval analysis to directly and more accurately compute difference in the values of neurons of the two networks from the input layer to output layer.  ... 
arXiv:2007.10135v1 fatcat:f5se22tlvje6ppgo2chdxjmmgq

Verification of Neural Network Behaviour: Formal Guarantees for Power System Applications [article]

Andreas Venzke, Spyros Chatzivasileiadis
2020 arXiv   pre-print
Such methods have the potential to build the missing trust of power system operators on neural networks, and unlock a series of new applications in power systems.  ...  Developing a rigorous framework based on mixed integer linear programming, our methods can determine the range of inputs that neural networks classify as safe or unsafe, and are able to systematically  ...  As an example, neural networks can be used to predict the security margin of the power systems (e.g. the damping ratio of the least damped mode in small signal stability analysis).  ... 
arXiv:1910.01624v3 fatcat:p4knzoxu3ffb7a6jlv55eafp2i

On Training Robust PDF Malware Classifiers [article]

Yizheng Chen, Shiqi Wang, Dongdong She, Suman Jana
2019 arXiv   pre-print
We demonstrate how the worst-case behavior of a malware classifier with respect to specific robustness properties can be formally verified.  ...  Furthermore, we find that training classifiers that satisfy formally verified robustness properties can increase the evasion cost of unbounded (i.e., not bounded by the robustness properties) attackers  ...  Different methods have been used to formally verify the robustness of neural networks over input regions [20, 23, 24, 31, 34, 37, 44] , such as abstract transformations [25] , symbolic interval analysis  ... 
arXiv:1904.03542v2 fatcat:p36c6kaxwzavbedcbbkaqybqyy

Feature Extraction Functions for Neural Logic Rule Learning [article]

Shashank Gupta, Antonio Robles-Kelly, Mohamed Reda Bouadjenek
2021 arXiv   pre-print
Combining symbolic human knowledge with neural networks provides a rule-based ante-hoc explanation of the output.  ...  In this paper, we propose feature extracting functions for integrating human knowledge abstracted as logic rules into the predictive behavior of a neural network.  ...  A recent paper [9] gives a detailed analysis on the methodology used in [7] , comparing its performance to other neural symbolic methods and arguing that it is not very effective in transferring knowledge  ... 
arXiv:2008.06326v4 fatcat:5xuiulpwofeh5ig5c3co6ppwbm
« Previous Showing results 1 — 15 out of 5,506 results