Scientific Meeting 6 May 2022

Chenglu Jin: PwoP Intrusion-Tolerant and Privacy-Preserving Sensor Fusion; Timo Deist: Multi-objective machine learning to predict Pareto fronts

6 May 2022 from 1 p.m. to 6 May 2022 2 p.m. CEST (GMT+0200)
Euler room (Z009)

13:00 - 13:30 Chenglu Jin (Computer Security), PwoP: Intrusion-Tolerant and Privacy-Preserving Sensor Fusion

We design and implement, PwoP, an efficient and scalable system for intrusion-tolerant and privacy-preserving multi-sensor fusion. PwoP develops and unifies techniques from dependable distributed systems and modern cryptography, and in contrast to prior works, can 1) provably defend against pollution attacks where some malicious sensors lie about their values to sway the final result, and 2) perform within the computation and bandwidth limitations of cyber-physical systems. PwoP is flexible and extensible, covering a variety of application scenarios. We demonstrate the practicality of our system using Raspberry Pi Zero W, and we show that PwoP is efficient in both failure-free and failure scenarios.

13:30 - 14:00 Timo Deist (Life Sciences and Health), Multi-objective machine learning to predict Pareto fronts

Real-world problems are often multi-objective with decision-makers unable to specify a priori which trade-off between the conflicting objectives is
preferable. Intuitively, building machine learning solutions in such cases would entail providing multiple predictions that span and uniformly cover
the Pareto front of all optimal trade-off solutions. We propose a novel approach for multi-objective training of neural networks to approximate the
Pareto front during inference. We also illustrate why training processes to approximate Pareto fronts need to optimize on fronts of individual training samples instead of on only the front of average losses. Experiments on three multi-objective problems (multi-objective regression, multi-observer image segmentation, neural style transfer) show that our approach returns outputs that are well-spread across different trade-offs on the approximated Pareto front without requiring the trade-off vectors to be specified a priori.