Internationally regarded as a leading researcher in the field of machine learning, Bernhard Schölkopf was the featured speaker for this year’s Courant Lectures in April, giving two talks: “Statistical and Causal Learning” and “Inference of Cause and Effect.” Currently on sabbatical from the Max Planck Institute for Intelligent Systems in Tübingen, Germany, where he is Director of the Department of Empirical Inference, Professor Schölkopf is spending several months in New York at the Courant Institute of Mathematical Sciences. Although the main reason for his sabbatical is to conduct research, he will also give a few lectures. In fact, he taught one class of a computer vision course at the Courant Institute in May.
Though it was certainly an honor to present this year’s Courant Lectures, Professor Schölkopf admits that it was also quite intimidating, once he looked at the list of past speakers. “The very first one was Eugene Wigner [1959] who gave a very famous talk, called ‘The Unreasonable Effectiveness of Mathematics in the Natural Sciences,’” he says. “I had read that before. So that was quite scary.” Since the first day’s lecture was meant to be of broad interest, he tried to make it very understandable to those of different mathematical backgrounds. “The field can be so rich and specialized that you can easily be lost if you have a different mathematical background.”
Machine learning finds regularities in data, then predicts based upon these regularities
Traditionally in science, Professor Schölkopf says, you make some measurements or experiments and try to come up with some comprehensible law or theory that describes these measurements.
In machine learning, the story is a bit different. “We can find regularities in data that humans cannot find,” he explains. “These regularities may not be simple in mathematical terms, because you are looking at systems that are very complex, maybe very high dimensional. We have to look at many quantities at the same time. We have to observe the system for a while, and measure many quantities. Then we put them into a learning algorithm that will end up giving us a complicated mathematical expression.”
How machine learning relates to data science
Given the fact that today’s systems work on huge data sets, possessing these data sets gives you a great deal of power, according to Professor Schölkopf. “Rob Fergus here at Courant is a computer vision guy, including image recognition systems,” he states. “He now has probably the best object recognition system in the world. It is machine-learning based, but he doesn’t have the largest data sets to train it on. These are owned by companies like Google, which records your clicks every time you do an image search. It has millions of images for all objects and categories that people search for, and of course these are not public. So it’s not just a question of who has the best algorithms to do this kind of learning, but also who has all this data. It raises a lot of interesting questions about data science.”
In addition, he says, with computers and machine learning algorithms, it’s now possible to predict using large data sets. “That’s why data is suddenly so important. If you want to do something intelligent with data, to extract knowledge that will be useful, it’s hard to do that without machine learning or statistics.”
NYU’s launch of the Initiative for Data Science and Statistics and the Center for Data Science
“It’s important to connect data to the actual sciences,” says Professor Schölkopf, “and what’s really nice about the Center for Data Science is that it involves not just machine learning people but also people from the sciences, like David Hogg in NYU’s Physics Department,” says. “He is one of my three hosts here at Courant, along with Rob Fergus and Yann LeCun, Director of the new Center for Data Science.” Coincidentally, Professor Schölkopf worked on the major part of his Ph.D. at Bell Labs in New Jersey in a department headed by Yann, almost 20 years ago. “It is nice to come back to a place headed by Yann,” he adds.
Schölkopf believes that machine learning may allow us to discover, or at least predict, properties of the world that have evaded simple mathematical descriptions – in biology, for instance. “While these regularities may not be comprehensible, we hope that there may be comprehensible methods for learning or inferring them,” he says. “That’s why this field aspires to be called ‘data science,’ and NYU is pioneering this area both in computer science and in the natural sciences.”
By M.L. Ball
To read the entire interview with Bernhard Schölkopf from which this story was written, click here.