In this occasional series, we ask instructors to discuss how they engage students in the great questions of our time.
Q. In a nutshell, what is this class about?
A. Biologically Inspired AI is a class focused on an emerging subfield of artificial intelligence that seeks to leverage insight from neuroscience to make computers more flexible and robust. Machine learning is seemingly everywhere these days, from Siri to Gmail’s Smart Compose. However, as “intelligent” as these systems might seem, their capabilities are actually extremely narrow. An app like Siri can only answer questions which resemble those it was trained on. And, more importantly, if we update these systems to allow them to perform a new task, they will likely lose the ability to perform old tasks.
Many research groups, in both universities and large tech companies, are looking to neuroscience to address these shortcomings. In short, researchers want to incorporate additional properties of the brain to make artificial neural networks behave more like the real thing.
In this course, we first review some machine learning and neuroscience basics, then look at some of these emerging techniques in detail.
Q. What are some goals you set for your students?
A. To be successful in this class, students must themselves be flexible and willing to learn many different subjects. Bio-inspired AI requires some working knowledge of machine learning, neuroscience, a little bit of biochemistry, probability, statistics, linear algebra, software development, etc. No one can be an expert in all these fields, so a key practical takeaway of this course is to learn how to learn about new subjects. My goal is for students to become independent learners who can cobble together multiple sources of information to arrive at a solution.
Q. What can students expect to get out of this class?
A. My main objective is for students to understand the relationship between biological and artificial neural networks—in what ways they are similar and in what ways they are different—and to understand how these differences affect the types of tasks that they can or cannot do well. At the end of course, students should understand the basic mathematical underpinnings of these systems and the corresponding biology that they’re trying to model. In addition, students are expected to implement machine learning algorithms and apply them on data.
I believe it’s vital for students to learn how to work in a field, rather than just learn about it. Data scientists, software engineers, researchers. etc. implement, apply, and extend existing methods. They don’t just read about them.
Q. How did you become interested in teaching this subject?
A. I first became fascinated with the relationship between brains and machines as an undergraduate student. I chose computer vision as my research area for my Ph.D. (my dissertation was on developing algorithms for automatically analyzing retinal vessels). In addition, I completed a graduate certificate in cognitive neuroscience during my Ph.D. to better understand how the brain works.
I currently have a grant from the National Science Foundation (NSF) in which I am studying how to combine artificial neural networks with neuromodulation, a form of chemical signaling that the brain uses to regulate neural activity. This class is a way for me to share what I know about this growing field with students.