Global ETD Search
Search theses and dissertations gathered from participating repositories worldwide. Every result links back to the library that holds it. No account is needed.
Results
Showing 1 to 20 of 50 for “"Backdoors"”.
-
Backdoors in Satisfiability Problems
… solve large practical instances. The notion of backdoors has been introduced to capture structural properties of instances. Backdoors are a set of variables for which there exists some value assignment that leads to a polynomial-time solvable sub-problem. I address in this thesis the problem of …
-
Backdoors in contemporary artificial intelligence: the good and the bad
… aspects of backdoor attacks. In particular, backdoors in the computer vision domain are often implemented as pixel patches. This form lends itself well to serving asa watermark for copyright protection—an increasingly important application in the era of generative AI. Building on this idea, …
-
Understanding and Enhancing CDCL-based SAT Solvers
… such as community structure, treewidth, and backdoors. We investigate how these parameters correlate with CDCL solving time, and whether they can effectively be used to distinguish formulas from different domains. We demonstrate how these measures can be used as a means to understand the …
-
Understanding and Extending Backdoor Attacks on Deep Neural Networks: From Behavioural and Mechanistic Analysis to Concept-Level Triggers
… accuracy, these metrics do not fully explain how backdoors change model behaviour, how defenses repair models, or what forms triggers can take. This thesis studies backdoor attacks from an analysis-oriented perspective. First, we ana-lyze backdoor attacks on image classification models through …
-
Exploiting Structure In Combinatorial Problems With Applications In Computational Sustainability
… or mixed-integer linear programs contain small backdoors. We study backdoors both theoretically and empirically and characterize important tradeoffs between the computational complexity of finding backdoors and their effectiveness in capturing problem structure succinctly. 2) We contribute …
-
Why Certified IoT Devices Remain Untrustworthy
… showing Internet connected devices contain backdoors of various origins whether they be placed intentionally or are discovered as vulnerabilities. These types of devices are of concern because they are critical for Internet access, are deployed in mass numbers, and would have a significant …
-
Backdoor in AI: Algorithms, Attacks, and Defenses
… strategy. Third, we actively explore leveraging backdoors for socially beneficial applications. We demonstrate that backdoors can be used for watermarking valuable assets within the deep learning pipeline. We focused on using backdoors as watermarks to protect data, models, and APIs. To monitor …
-
Towards Safe and Secure Generative AI Systems
… patching defense called Mudjacking to remove backdoors from foundation models after an attack is detected. Because a compromised foundation model can undermine every intelligent application built on it, Mudjacking is crucial for safeguarding the broader ecosystem of generative AI. In …
-
(Un)Trustworthy Machine Learning
… devices. This introduces vulnerabilities such as backdoors, i.e., unrelated tasks that the model may unintentionally learn when an adversary controls parts of the training data or pipeline. In this thesis, we identify new threats to ML models and propose approaches that balance security, accuracy, …
-
Parameterized complexity in graph theory, logic and machine learning
… standard graph parameters, this work utilises ``backdoors'', which are sets of variables whose removal yields tractable classes like Quantified 2-CNF. By introducing ``Enhanced Backdoors'', the research bridges the gap between syntactic restrictions and structural graph properties, significantly …
-
Crumpled and Abraded Encryption: Implementation and Provably Secure Construction
… scheme difficult to analyze and allows possible backdoors. In this thesis, we introduce a new abrasion construction which uses hash functions to avoid the problems with the current abrasion construction. In addition, we present a proof-of-concept for using crumpled encryption on an email server.
-
The Interdependence of Things. Digital product interdependencies and European cybersecurity law
… NSA, a US intelligence agency created virtual ‘backdoors’ in some products before shipping them to the customers. In essence, modern cybersecurity threats are similar to the Trojan horse: they lie hidden beneath layers of components that are largely unknown to the average consumer. The European …
-
Backdoor detection systems for embedded devices
… to discover previously undocumented real-world backdoors.
-
Constraint Satisfaction Techniques for Combinatorial Problems
… results for problems about backbones and backdoors of Boolean formulas. We consider applications of constraint satisfaction techniques to problems in graph arrowing (specifically in Ramsey and Folkman theory) and computational social choice. Our contributions show how applying constraint …
-
Fortifying robustness: unveiling the intricacies of training and inference vulnerabilities in centralized and federated neural networks
… the practical black-box defense against Trojan backdoors has failed to achieve state-of-the-art performance. More concerning is the limited exploration of these vulnerabilities within the context of cooperative attack or Federated learning, leaving NN classifiers exposed to unknown risks. This …
-
Classifying malicious windows executables using anomaly based detection
… Maiware, which includes Viruses, Worms, Trojans, Backdoors, Root-kits, Spyware and Exploits. Anomaly detection is technique which builds a statistical profile of the normal and malicious data and classifies unseen data based on these two profiles. A detection system is presented here which is …
-
Detecting Compromised Hardware Integrity with Machine Learning in Multicore Processors
… hardware (including hardware trojans), hardware backdoors, IP theft, and counterfeit hardware. This thesis will focus on the issue of counterfeit hardware. As the complexity of modern day hardware has increased, detecting counterfeits has become increasingly difficult. Current methods of physical …
-
Toward Trustworthy Machine Learning with Blackbox and Whitebox Methods
… a small portion of training data to inject backdoors. The proposed solutions include a neural-guided sticker reverse engineering technique and an ensemble training method based on a novel backdoor detection code. While being universal to all types of ML systems, the blackbox methods also …
Page 1 of 3