back

SING: Improving the efficiency of Muli-Party Computation protocol assignment using Neural Networks

Master Thesis

Motivation

Multi-party computation protocols enable multiple parties to compute on their private data. There are multiple protocols available, the most common being GMW [5] and Yao's Garbled Circuits [6]. These protocols have different strengths and weaknesses, rendering the task of optimal protocol selection for a function NP-hard.

Recent works have focused on tackling the protocol assignment task through heuristics [1] or integer linear programming [2,4]. However, both of these approaches incur a high performance cost, with protocol assignment taking ~40 minutes for large circuits.

Goal

Instead of hard-coded heuristics or integer linear programming, the thesis will evaluate the suitability of machine learning models such as (deep) neural networks for protocol assignment. While the resulting assignments will not be globally optimal, they have the potential of being produced in a significantly faster manner.

Another direction is to explore different strategies for learning protocol assignment such as supervised, unsupervised, or semi-supervised learning. A suitable training and testing dataset is collected where applicable.

Finally, the performance of the new approach is compared with previous works.

Requirements

  • Programming skills in Python, C++ or Rust
  • At least basic knowledge of cryptography
  • Familiarity with machine learning in any common Python framework
  • High motivation + ability to work independently
  • Knowledge of the English language, Git, LaTeX, etc.

References

Supervisors