Pre-Memorial Day Kickoff

Hello Julia enthusiasts! Just like anything worthwhile and enduring, it all starts with humble beginnings. This is Julia5280's kickoff event and humble start. [Note: Due to COVID-19, this will be an online event.]

---

Agenda:

• Quick Intros (10 min)
• Housekeeping + Introduce Presenter (3 min)
• Main Talk - "How Julia makes new decision-making AI possible" by Zach Sunberg, PhD (20 min)
• Q&A (10 min)
• Sponsorships for future online and/or physical meetup events (2 min)
• Virtual Networking (30-45 min)

---

How Julia makes new decision-making AI possible

Every aspect of human life involves making decisions in the face of uncertainty, from steering a car to prescribing a cancer treatment. Artificial intelligence has the potential to help us safely improve our well being in these cases. The Markov decision process (MDP) is a framework for expressing a sequence of decisions as an optimization problem, and the partially observable Markov decision process (POMDP) provides a way to model both aleatory and epistemic uncertainty. The techniques used to solve these problems are often called reinforcement learning or dynamic programming.

Research in this area is challenging for several reasons. First, solving large MDPs and even moderately-sized POMDPs is computationally difficult, meaning that every bit of speed is needed. Second, the algorithms used to solve MDPs and POMDPs are often complex, approximate, and tailored to specific classes of problems, so code clarity and modularity is vital. Third, the variety of problems that can be expressed as MDPs and POMDPs is huge, requiring software to have the utmost flexibility.

Julia is an ideal language for studying these problems because it combines speed, clarity, and flexibility. We have implemented the POMDPs.jl framework to help us study MDPs and POMDPs. Several important algorithmic advances are the direct result of the ease with which we can explore the space of algorithms with Julia, and we are excited to investigate new directions in the future with additional features of Julia such as automatic differentiation.

About Zach Sunberg, PhD

Zach Sunberg is an Assistant Professor in the Smead Aerospace Engineering Sciences Department at the University of Colorado Boulder. He received Bachelor's and Masters degrees in Aerospace Engineering at Texas A&M University before completing a PhD in Aeronautics and Astronautics at Stanford University with a focus on safe and efficient decision making for autonomous vehicles. Before coming to Boulder he was a postdoctoral scholar at the University of California, Berkeley. His current research focus is on decision-making algorithms that robustly handle various types of uncertainty in the continuous domains encountered by real vehicles. He is a native Coloradan and avid skier.