A Look Into the Future of Machine Learning

By on

Sunrise at Millabedda, Sri Lankaby Angela Guess

David Karger, a professor of Computer Science at MIT, recently wrote on Forbes, “I think that the long-term future of machine learning is very bright (and that we will ultimately solve AI, although that’s a separate issue from ML).   Machine learning is already an incredibly powerful tool that can do a surprisingly good job of solving really hard classification problems. But, as a research area it doesn’t really appeal to me.  As a theoretician, I tend to be drawn to (i) solving specific problems rather than coming up with general paradigms and (ii) devising solutions that may be tricky to think up, but that are intuitive and easy to understand and explain.  So if you see my (limited) past work on machine learning, you’ll see I’ve tackled problems like ‘how do we set parameters of a simple weighted-linear classifier in order to retrieve relevant documents’.  The field of machine learning has moved way beyond that kind of question, developing incredibly powerful and general algorithms that rely on deep and sophisticated math.  I’m not so interested in doing that kind of abstract work.”

Karger goes on, “Another turn-off for me is that with these incredibly powerful algorithms, you can solve really hard problems, but while the computer knows the answer it just works like magic.  You don’t really know *why* that’s the answer.  A concrete example of this is the difference between so-called ‘generative’ and ‘discriminative’ models.  Generative models in machine learning posit that there is some underlying (generally random) process that is generating the data you are observing, and aim to use the data to infer the parameters of that underlying process, which then lets you classify the data. Discriminative models don’t care how the data is being generated; they just figure out a formula that effectively distinguishes the different classes of data. In my mind, if you succeed in solving a generative model, you have ‘understood’ the data and the problem.”

Read more here.

We use technologies such as cookies to understand how you use our site and to provide a better user experience. This includes personalizing content, using analytics and improving site operations. We may share your information about your use of our site with third parties in accordance with our Privacy Policy. You can change your cookie settings as described here at any time, but parts of our site may not function correctly without them. By continuing to use our site, you agree that we can save cookies on your device, unless you have disabled cookies.
I Accept