Assistant Professor Josiah Hanna comes to UW-Madison Computer Sciences to focus on machine learning and robotics

Josiah Hanna comes to UW-Madison from earning his PhD at the University of Texas-Austin. Hanna works at the intersection of machine learning and robotics, focusing on reinforcement learning (RL). RL “allows computers and robots to learn from trial and error interaction how to successfully complete tasks rather than being explicitly programmed how to complete the task.” Hanna hopes his students will come away from his classes understanding that machine learning and artificial intelligence can help solve exciting real-world problems, but there is still much research to be done. 

Hometown: Lexington, Kentucky

Educational/professional background: 
Ph.D. in Computer Science from the University of Texas at Austin, B.S. in Computer Science and Mathematics from the University of Kentucky.

How did you get into your field of research?
I started with undergraduate research in artificial intelligence (AI). I was able to learn a lot about the different areas of AI research and then decide that reinforcement learning (RL) and robotics was what excited me the most and what I wanted to keep studying for my PhD. A lot of AI research focuses on isolated intelligent capabilities, for example, how an intelligent agent should detect objects or conduct logical reasoning. I find RL and robotics to be particularly exciting because they focus on building complete intelligent agents that have to both understand their surrounding world and then choose actions to accomplish the goals set by their designers. 

Could you please describe your area of focus? 
I work at the intersection of machine learning and robotics. Specifically I focus on a type of machine learning called reinforcement learning (RL). RL allows computers and robots to learn from trial and error interaction how to successfully complete tasks rather than being explicitly programmed how to complete the task. 

What main issue do you address or problem do you seek to solve in your work? 
The main challenge my research addresses is how we can design RL algorithms that learn quickly, without long periods of learning before a task can be completed satisfactory. Right now, an RL algorithm can be used to allow a computer to play games at super-human level. However, learning is only possible because it’s relatively easy and quick for the computer to play millions of games against itself to figure out the optimal way to play. The same type of learning can’t easily scale to robots that have to act in the physical world in real time.

What’s one thing you hope students who take a class with you will come away with? 
A lot of exciting real world problems can be solved with machine learning and artificial intelligence but, despite a lot of hype, there is still much research to be done towards building complete intelligent machines such as the robots from science fiction.

What attracted you to UW-Madison? 
UW—Madison has a great combination of research excellence while being a supportive and friendly environment. You can tell from the outside that people respect one another and are working to make it a place where everyone can succeed. And it’s also in a fun city to live in!

What was your first visit to campus like? 
My first visit was in February 2019. I was living in Texas at the time and was really impressed by how much outside activity was going on considering how much snow there was!

What are you looking forward to doing or experiencing in Madison?
I love the city’s commitment to maintaining nature and parks. I’m looking forward to continue exploring the different parks and trails around the city. 

Do you feel your work relates in any way to the Wisconsin Idea? If so, please describe how. 
Reinforcement learning has the potential to enable more intelligent decision-making in fields from education to robotics to health care and thus to provide many positive benefits for society. The aim of my work is to make reinforcement learning easier to use for real world applications and thus to realize this potential for the wider world.

Please tell us about something you’re working on in layperson’s terms, so that non-computer scientists at UW-Madison and the general public can understand what you’re passionate about.
I’m working on allowing robots to learn skills in simulated environments (virtual worlds) and then transfer the learned skills the real, physical world. The advantage of learning this way is that we can simulate large numbers of robots (potentially 100s of identical robots) at the same time, and they can share experience to acquire new skills much faster than learning with a single robot in the real world. The challenge is that simulated environments can’t capture all the complexity of the real world and so skills the robots learn may fail to generalize to the real world. For example, in prior work, we trained a robot to walk fast in a simulated environment, and it learned to walk really fast in simulation but tripped and fell in reality. We’re working on new approaches that allow robots to learn in simulated environments and the skills they learn to successfully transfer to the real world.

Hobbies/other interests:
I like being outside and being active so I enjoy hiking, running, and biking. I also like reading and traveling.