Computer Science Colloquia
Friday, May 2, 2014
Advisor: David Evans
Attending Faculty: Westley Weimer (Chair), abhi shelat, Mohammad Mahmoody, and Denis Nekipelov (Minor Representative)
2:00 PM, Rice Hall, Rm. 242
PhD Proposal Presentation
Abstractions for Data Oblivious Programs
While many recent papers have demonstrated the feasibility of secure computation for various interesting applications, such techniques have not yet been widely adopted outside of the research community. In the thesis proposed here, we try to reduce one aspect of this entry barrier: software abstractions. We motivate the problem by showing how secure computation necessarily requires redesigning of even simple software abstractions such as language control structures and data structures. First, we propose a new language that can be easily extended by other researchers for purposes of their investigations. Then, we propose new constructions for common data structures that are efficient in this execution model. Finally, we propose to develop new optimizations for ORAM structures to enable faster computations in the RAM model. Our preliminary investigations are already showing promising results. We have implemented a prototype compiler for our new language that provides significantly higher flexibility compared to existing systems. We demonstrate this flexibility by showing that our language allows implementation of various library-based features that have, in the past, always used compiler modifications in other languages. We have also shown constructions of data structures that can provide over 10x speed improvement even on small data sizes.