Explore a groundbreaking framework for parallel secure computation in this IEEE Symposium presentation. Delve into GraphSC, a system designed to simplify secure code writing for non-cryptography experts, introduce parallelism to secure implementations, and maintain data privacy through obliviousness. Learn how this innovative approach enables efficient, secure execution of graph-based algorithms, including complex data mining and machine learning tasks, on large datasets. Discover the framework's ability to process graph-based algorithms with only a small logarithmic overhead compared to non-secure parallel versions. Examine the practical applications of GraphSC in big data analysis, including a secure matrix factorization implementation capable of processing 1 million ratings in 13 hours. Gain insights into achieving parallelism, ensuring obliviousness in graph-parallel algorithms, and implementing the oblivious gather-key trick. Analyze experimental results, scalability, and potential applications across data centers.
Overview
Syllabus
Intro
Companies Computing on Private Data
Key Challenges
Key Contributions
Achieving Parallelism
Obliviousness of Graph-parallel Algorithms
Oblivious Gather-Key Trick
Complexity of Our Algorithms
Experimental Setup
Key Evaluation Results
Running at Scale
Conclusion
Across Data Centers
Taught by
IEEE Symposium on Security and Privacy
