SearcharxivSearch

arXiv subjects

Andrew M. Webb

Publications and source records attributed to Andrew M. Webb.

2 recordsLinked to original sources

SoK: Come Together -- Unifying Security, Information Theory, and Cognition for a Mixed Reality Deception Attack Ontology & Analysis Framework

We present a primary attack ontology and analysis framework for deception attacks in Mixed Reality (MR). This is achieved through multidisciplinary Systematization of Knowledge (SoK), integrating concepts from MR security, information theory, and cognition. While MR grows in popularity, it presents many cybersecurity challenges, particularly concerning deception attacks and their effects on humans. In this paper, we use the Borden-Kopp model of deception to develop a comprehensive ontology of MR deception attacks. Further, we derive two models to assess impact of MR deception attacks on information communication and decision-making. The first, an information-theoretic model, mathematically formalizes the effects of attacks on information communication. The second, a decision-making model, details the effects of attacks on interlaced cognitive processes. Using our ontology and models, we establish the MR Deception Analysis Framework (DAF) to assess the effects of MR deception attacks on information channels, perception, and attention. Our SoK uncovers five key findings for research and practice and identifies five research gaps to guide future work.

cs.CR

To Ensemble or Not Ensemble: When does End-To-End Training Fail?

End-to-End training (E2E) is becoming more and more popular to train complex Deep Network architectures. An interesting question is whether this trend will continue-are there any clear failure cases for E2E training? We study this question in depth, for the specific case of E2E training an ensemble of networks. Our strategy is to blend the gradient smoothly in between two extremes: from independent training of the networks, up to to full E2E training. We find clear failure cases, where over-parameterized models cannot be trained E2E. A surprising result is that the optimum can sometimes lie in between the two, neither an ensemble or an E2E system. The work also uncovers links to Dropout, and raises questions around the nature of ensemble diversity and multi-branch networks.

stat.ML