Search

Oblivious inference is the task of outsourcing a ML model, like neural-networks, without disclosing critical and sensitive information, like the model's parameters. One of the most prominent solutions for secure oblivious inference is based on a powerful cryptographic tools, like Homomorphic Encryption (HE) and/or multi-party computation (MPC). Even though the implementation of oblivious inference systems schemes has impressively improved the last decade, there are still significant limitations on the ML models that they can practically implement. Especially when both the ML model and the input data's confidentiality must be protected. In this paper, we introduce the notion of partially oblivious inference. We empirically show that for neural network models, like CNNs, some information leakage can be acceptable. We therefore propose a novel trade-off between security and efficiency. In our research, we investigate the impact on security and inference runtime performance from the CNN model's weights partial leakage. We experimentally demonstrate that in a CIFAR-10 network we can leak up to $80\%$ of the model's weights with practically no security impact, while the necessary HE-mutliplications are performed four times faster., P. Rizomiliotis, C. Diou, A. Triakosia, I. Kyrannas and K. Tserpes. Partially oblivious neural network inference. In Proceedings of the 19th International Conference on Security and Cryptography, SECRYPT (pp. 158-169), 2022

This the second deliverable of Work Package 2 tasks T2.1 (Tools and methods for secure data sharing), T2.2 (Trustworthiness of data flows), T2.3 (Machine learning-based cognitive security framework), T2.4 (Statistical Analytics and Machine- / Deep-Learning on shared data), T2.5 (Distributed anomaly detection for Industrial IoT) and T2.6 (Workflow-driven security for supply chain and compliance in manufacturing) related to the 1st version of integrated platform of the project. It describes and demonstrates the various technologies that form the COLLABS Level-3 security package for secure digital supply networks.&nbsp

In this paper we experimentally evaluate a physical unclonable function based on a polymer optical waveguide, as a time-invariant, replication-resilient, source of entropy. The elevated physical unclonability of our implementation is combined with spatial light modulation and post processing techniques, thus allowing the deterministic generation of an exponentially large pool of unpredictable responses. The quality of the generated numbers is validated through NIST/DIEHARD(ER) suites, whereas the overall security of the scheme is benchmarked assuming attackers with elevated privileges in terms of system access. Finally, based on the demonstrated key features, we present and analyze a mutual authentication implementation scenario which is fully compatible with state-of-the-art commercial Internet-Of-Things architectures, Comment: 14 pages, 7 figures, 4 tables

International audience; In this chapter we present an innovative approach towards the design and application of Security and Dependability (S&D) solutions for Web services and service-based workflows. Recently, several standards have been published that prescribe S&D solutions for Web services, e.g. OASIS WS-Security. However,the application of these solutions in specific contexts has been proven problematic. We propose a new framework for the application of such solutions based on the SERENITY S&D Pattern concept. An S&D Pattern comprises all the necessary information for the implementation, verification, deployment, and active monitoring of an S&D Solution. Thus, system developers may rely on proven solutions that are dynamically deployed and monitored by the Serenity Runtime Framework. Finally, we further extend this approach to cover the case of executable workflows which are realised through the orchestration of Web services.