Smart contracts have witnessed widespread adoption across various industries since the introduction of Ethereum. Also, smart contracts’ vulnerabilities have gradually attracted academic attention, e.g., Oyente was de...
详细信息
Deep Neural Networks (DNNs) have found successful applications in various non-safety-critical domains. However, given the inherent lack of interpretability in DNNs, ensuring their prediction accuracy through robustnes...
详细信息
ISBN:
(纸本)9783031664557;9783031664564
Deep Neural Networks (DNNs) have found successful applications in various non-safety-critical domains. However, given the inherent lack of interpretability in DNNs, ensuring their prediction accuracy through robustness verification becomes imperative before deploying them in safety-critical applications. Neural Network Verification (NNV) approaches can broadly be categorized into exact and approximate solutions. Exact solutions are complete but time-consuming, making them unsuitable for large network architectures. In contrast, approximate solutions, aided by abstraction techniques, can handle larger networks, although they may be incomplete. this paper introduces AccMILP, an approach that leverages abstraction to transform NNV problems into Mixed Integer Linear Programming (MILP) problems. AccMILP considers the impact of individual neurons on target labels in DNNs and combines various relaxation methods to reduce the size of NNV models while ensuring verification accuracy. the experimental results indicate that AccMILP can reduce the size of the verification model by approximately 30% and decrease the solution time by at least 80% while maintaining performance equal to or greater than 60% of MIPVerify. In other words, AccMILP is well-suited for the verification of large-scale DNNs.
In this paper, we consider the problem of prediction of Radio Link Failures (RLF) in flying ad hoc networks (FANETs). Many environmental factors that influence the quality of radio wave propagation are dynamic, and th...
详细信息
Legal contracts have been used for millennia around the world as a foundation for business transactions. Smart con-tracts are cyber-physical systemsthat deploy Internet-of-things (IoT) and blockchain technologies to ...
详细信息
Cybersecurity is increasingly threatened by the growing prevalence of spyware and other malicious software intended to enter computersystems and modify data. this research paper proposes a ground- breaking amalgamati...
详细信息
software-Defined Networking (SDN) enhances flexibility, scalability, and innovation by decoupling the control plane from the data plane, managed through streamlined controller operations. However, Distributed Denial o...
详细信息
the bumper beam assembly of an automobile is crucial for absorbing impact energy and shielding occupants from front and rear collisions. Crash testing is a type of destructive testing that is typically used to verify ...
详细信息
the perception of the autonomous driving software of the FS223, a low-level sensor fusion of Lidar and Camera data requires the use of a neural network for image classification. To keep the neural network up to date w...
详细信息
In view of the problems of cold start and data interaction in recommendation systems, and most current recommendation algorithms ignore the diversity of data types, the combination of multimodal data and knowledge gra...
详细信息
this article reviews earlier articles on the topic of residuality theory and places residuality theory in the context of the complexity sciences, relating the major concepts of residuality theory in terms of Kauffman ...
详细信息
暂无评论