Chair: Raphael Wimmer
A Living Lab Architecture for Reproducible Shared Task Experimentation
Timo Breuer, Philipp Schaer
TH Köln – University of Applied Sciences, Germany
No existing evaluation infrastructure for shared tasks currently supports both reproducible on- and offline experiments. In this work, we present an architecture that ties together both types of experiments with a focus on reproducibility. The readers are provided with a technical description of the infrastructure and details of how to contribute their own experiments to upcoming evaluation tasks.
Keywords: reproducibility; evaluation infrastructure; shared tasks
procd: A privacy-preserving robust implementation to discover contacts in social networks
Fabian Deifuß1, Cornelius Ihle1, Moritz Schubotz2, Bela Gipp1
1University of Wuppertal, Germany; 2FIZ Karlsruhe, Germany
Current instant messengers store the users’ phone book contacts typically unencrypted or hashed on a central server. In case of a server’s corruption, all contacts are either directly available in plaintext or can be unmasked using a simple dictionary attack. To solve this problem, we present procd [pʁoːst] a python implementation for privacy preserving contact discovery. procd is a trustless solution that requires neither plaintext numbers nor hashes of single phone numbers to retrieve contacts. Instead, we transfer hashed combinations of multiple phone numbers, which increases the effort for dictionary attacks to an unfeasible level using today’s hardware.
Keywords: private contact discovery; private set intersection; secure multi-party computation; private information retrieval
Event Timeslots (1)