Dimitri Zafirov | How Do Machines Learn?

A scholar with a background in finance and economics, Dimitri Zafirov is working toward his doctorate in accounting at the University of California, Los Angeles (UCLA). Dimitri Zafirov’s research interests include the potential applications of machine learning.

When a machine proves capable of learning, it means that it can perform tasks that it was not originally programmed to perform. Such machines develop “skills” as a result of the datasets that they examine, and they do so with minimal or even no intervention of a human.

This differs from the traditional computer programming model. In that model, a human makes a program. The human runs the program on a computer to make sense of data also input by a human. The computer then produces useful output.

To train a computer to learn, humans give the computer data as well as useful output related to that data. The computer then runs through that data and output, designing its own program in the process. Humans can then evaluate that machine-designed program to determine its effectiveness in performing the task at hand.

Machine learning has staggering potential to transform the world. For example, computers may one day prove better than humans at learning from and accurately interpreting diagnostic data, such as those produced by body scans. They may also, by means of the vast quantity of data they can process, prove better than humans at making accurate predictions regarding financial risk.