ETH Zurich : Computer Science : Pervasive Computing : Distributed Systems : Education : Student Projects : Abstract

Privacy-Preserving Decentralized Analytics (M)

Status: Offen

Motivation

As of today, it is often common practice to pool data to centralized servers to conduct large-scale analytics over data from multiple sources and users. Therefore, most data processing pipelines are optimized to work in centralized settings. Recently, we have seen efforts that explore alternative mechanisms where storage and data processing are decoupled, federated learning is an embodiment of such efforts. There are many reasons that drove the recent interest in this computing paradigm, to mention a few; the unprecedented scale of generated data and the need to optimize data transfer over WAN, privacy concerns that arise from storing sensitive data remotely, the emergence of edge computing, and data sovereignty/transfer regulations. However, we have several distinct technical, privacy, and security challenges that need to be carefully addressed in the emerging paradigm of decentralized analytics.

Your Role

We have several theses opportunities in this project. Particularly we are a looking for students who are interested in joining efforts in building a framework for private decentralized analytics or work on designing privacy-preserving mechanisms for decentralized analytics.
Recommended readings:

• "Security and Privacy of Machine Learning", Overview Talk by Ian Goodfellow
• Blog: "Federated Learning: Collaborative Machine Learning without Centralized Training Data"
• Exploiting Unintended Feature Leakage in Collaborative Learning
• Communication-Efficient Learning of Deep Networks from Decentralized Data

Requirements:

We are particularly interested in students with a background and research interests in at least one of the following areas: machine learning, systems, and security. The student is expected to have good experience with Java, C++, or python and be interested in working with new tools. We expect our students to be highly motivated to work on their topics and to cooperate with their supervisors regularly to discuss current progress and next steps.

What we offer:

This topic will give you the opportunity to learn and get involved in timely systems and security research problems. To speed up your learning curve we will support you with tutorials and how-tos. We offer you a great work atmosphere, which is both casual and challenging, motivated advisors, and a good coffee machine ;)

Contact