Prof. Benny Pinkas is a Professor in the Department of Computer Science. Together with his team, he studies the design and analysis of solutions to applied security problems using the methods of the theory of cryptography, with a focus on secure multi-party computation.
Pinkas and his team study computer security, privacy and cryptography, and in particular in the design of efficient security systems based on sound assumptions and solid proofs.
One of the most attractive contributions of modern cryptography is secure computation, which allows multiple participants, each with its own private input, to communicate without the help of any trusted party, and compute any function of their inputs without revealing any information about the inputs (except for the value of the function). A classic example of such a computation is the “millionaires problem,” in which two millionaires want to find out who is richer, without revealing their actual worth.
Thus far, secure computation techniques have rarely been applied in practice, and are typically considered to have mostly theoretical significance. Pinkas and his team aim to build tools that translate these theoretical results into practical applications. Their goal is to create secure computation solutions—which today are usually stated as mathematical theorems—that can be used by non-experts, similar to state-of-the-art tools for technologies such as public key encryption, linear programming, or data compression.To that end, they develop generic tools (essentially compilers) that translate functions, defined in a high-level language, to distributed programs that implement a secure computation of the defined functions.
In addition, the group works on the design of specialized and highly efficient solutions to key tasks that have the conflicting goals of respecting privacy and enabling the legitimate usage of data.
Together with his research group, Pinkas has developed a system for face identification that compares the faces of subjects with a database of registered faces. The identification is done in a secure way that protects both the privacy of the subjects and the confidentiality of the database. The project provides a new face identification algorithm—which is unique in having been specifically designed for usage in secure computation—that displays a performance comparable to those of state-of-the-art algorithms which lack any security guarantee. The algorithm is robust to different viewing conditions, such as illumination, occlusions, and changes in appearance (such as wearing glasses).
Since the team’s goal is to run an actual system, they exerted considerable effort on optimizing the protocol and minimizing its online latency. The system, which implements a secure computation of the face identification protocol, can run in near real-time. The secure computation protocol performs a preprocessing of all public-key cryptographic operations. Its online performance therefore mainly depends on the speed of data communication and the team’s experiments have shown it to be extremely efficient.
A specific application of the system is in reducing the privacy impact of camera-based surveillance. It could be used in a setting that contains a server (with a set of suspects’ faces) and client machines (such as cameras that acquire images in public places). The system would run a secure computation of a face recognition algorithm and would identify if an image acquired by a client matches one of the suspects. Information would be revealed only if an appropriate match was found.