Advancing machine learning while protecting privacy
Throughout the last century, we created and perfected the art of computer programming to automate mundane tasks.
Conventional programs, however, still rely on hand-crafted rules, making programming a complicated and tedious task.
Suppose you were to develop an email spam filter. Using traditional programming, you would have to read through millions of emails to deduce rules that can filter out annoying junk mail more or less reliably.
Surely, there must be a more efficient way. And this is what machine learning, one of my areas of research, is about.
In essence, machine learning is a set of techniques for teaching computers how to learn predictive rules from data and turning these into programs automatically.
Now, machine learning is everywhere: face recognition on our mobile devices, voice assistants, recommendation systems, and soon, self-driving cars.
The more data that are available, the better machine learning works. At the same time though, the way data are collected and collated is a significant concern for personal privacy.
The purpose of technological progress is to solve problems, not to create them.
As an artificial intelligence and machine learning researcher, my focus is on making machine learning more secure and respectful of users’ privacy.
For example, we developed algorithms for storing face images so that personal information such as gender and age is concealed while these images retain their utility for face-matching applications.
Now, we are focusing solutions for other application areas of machine learning as well so that we can embrace technological progress without adverse side effects.