Cryptography with Noisy Data

Dr. Smith is a Post-doctoral Researcher at the Weizmann Institute of Science in Rehobot, Israel; he received his PHD from MIT in 2004
This talk presents a rigorous framework and new mathematical tools for
handling cryptographic keys which are inherently subject to errors.

As a concrete example, consider biometric password authentication,
where a user authenticates herself to a remote server using some
partially secret physical data, such as an iris scan or fingerprint.
Traditional authentication protocols assume that the user's "password"
is reproduced exactly each time it is read. (For example, they may
work by comparing a hash of the putative password to a hash of the
real password.) In contrast, biometric measurements are subject to
noise, and traditional solutions fail dramatically.

I will explain how to design noise-tolerant hash functions for

— turning noisy information into keys usable for any cryptographic
application, and, in particular,

— reliably and securely authenticating biometric data.

These results lead to new uses of randomness extractors in
cryptography, and new ways of quantifying the leakage of "useful"
information in a cryptographic protocol.