DCWHome

Coming from a theoretical computer science/cryptography background, I tend to find more appealing solutions that provide *mathematically provable assurances* that no significant privacy violation will occur. Such assurances should be robustly defined in a way that does not presume the methods and information available to parties attempting to extract private information from published data. In cryptography, we have been able to find such tools - the science and technology encryption is today sufficiently developed that many seemingly paradoxical tasks are widely used and trusted by individuals, corporations and governments (e.g., using your credit card on the web without the actual number exposed on any intermediate router, or backing up data on a remote server without revealing the data to the server's owner). Alas, in the field of data privacy and statistical analysis currently there seems to be quite a gap between the techniques and tools that provide such rigorous assurances, and the techniques and tools that are actually used for data analysis. I am yet unsure how this gap will be closed: perhaps we will have better techniques and tools with rigorous mathematical analysis, perhaps it will turn out that the tools currently used today have inherent privacy issues, or perhaps a mixture of both. I hope to learn more about this during this workshop.