How can we teach our machines to reason? Researchers in machine learning, signal processing, and information theory develop algorithms and computer-based procedures to mimic human perceptions, thoughts and actions. By looking at large amounts of data, learning methods are able to extract relevant information from video, photographs, audio, text, ultrasound, medical images, as well as a wide range of sensors and experimental measurements. But what is “information,” and how is it gathered, stored and processed? ECE researchers tackle these kinds of fundamental questions, applying mathematical reasoning, optimization procedures, and computational tools to engineering challenges. We seek technological solutions to problems in a variety of domains such as communications, medicine, the physical sciences, and the humanities.
Recurrent neural networks are the computing engines behind state-of-the-art applications from self-driving cars to speech recognition like Amazon’s Alexa. The behavior of these networks is challenging to characterize, but it can be visualized for small networks. This image, and the one at the top of this page, displays the behavior of a network, showing the way their output evolves by mapping their values in color. The result, a fractal structure, could help researchers better understand the behavior and characteristics of these kinds of networks. Photo credit: Professor Robert Nowak