Skip to main content

Embracing Uncertainty: The Morrison Group

Rebecca Morrison, an assistant professor of computer science at CU Boulder, focuses on the failures and uncertainties inherent in large computational models and how to create models when others fail. 

She and her group are also interested in how to create predictive models for datasets that are highly interrelated, but don't follow a traditional 'normal' distribution. 

Morrison believes that the key to a successful PhD is a healthy group culture. 

"I think there's sometimes a mentality in academia that you need to be stressed, but I try to set the example that it's okay to be okay, to get enough sleep and not be overworked. You can still be working hard enough and be happy and spend time with friends," she said. 

Let's learn more about the group's research from its members: 

Rileigh Bandy, 5th Year PhD

Rileigh Bandy, a fifth-year PhD student, is researching how to represent and understand the uncertainties in models which represent elements of the natural world. 

These physics-based models can help scientists predict weather patterns, tumor growth, stress patterns on bridges and more. 

"Generally we can divide those models into two camps," Bandy said. "There are white box systems of equations or black box models that just spit out outputs where we don't know how they're working. My research has been exploring how we can correct both of those models," she said.  

Though highly advanced, these models cannot hope to perfectly simulate the chaotic natural world. Understanding how uncertain the models are helps to improve them, compare them and create reasonable trust levels. 

Many physics-based models are safety-critical, such as weather forecasts for flight-paths or tumor-growth predictions for cancer treatment, and understanding them well is important to our everyday lives. 

Bandy said that there are many paths to a PhD. 

"To find your right path, focus on what you're most interested in. Try to find an advisor and collaborators that have common interests that you like to work with, because that's how you'll want to spend five-ish years doing it," she said. 

Teo Price-Broncucia, 5th year PhD student 

Teo Price-Broncucia, a fifth-year PhD student, is seeking ways to reduce the expense and code-base complexity of large scientific computer models. 

''One thing I'm working on is calibration methods to develop cheaper reduced models with slightly lower fidelity that hopefully retain a lot of the original models' usefulness," he said. 

Price-Broncucia explained that these models might be useful in education or industry, but also for science itself.  

"There are a lot of times you want to make lots of model runs, and that's difficult to do if it takes a full supercomputer three weeks to run a single model run," he said. 

He has also been working with the National Center for Atmospheric Research (NCAR) to understand how changes to the code of chaotic climate models impact what they output. 

"Really small changes to a chaotic model can result in totally different outputs. It's the butterfly effect. It's difficult to say if something meaningful changed or a single bit changed somewhere, and thus we got a different output, but really the model is the same," he said. 

Price-Broncucia believes the best way to pursue a PhD is sustainably, akin to a marathon, not a sprint. 

"You can't put your life on hold when you do a PhD. Research is a creative enterprise. You have better ideas when you're not totally sleep-deprived and stressed out and when you're able to have some space to think about things in a nice way," he said.

Ujas Shah, 1st year PhD Student

Ujas Shah's undergraduate education was in politics and economics before he started a master's degree in data science, which he has now transferred into a PhD. 

"My master's was the first time where I actually got into the sciences. I would say that we shouldn't be so afraid of switching fields," Shah said. 

Shah is focusing on graphical models for datasets that don't follow the classic bell-curve 'normal' distribution, but are still interlinked and affect one another. 

These datasets capture the intricacies of many layers of interconnected variables, such as complex company mergers or how animal species interact in a certain area. 

"We already have methods established for a normal distribution, but I am building on some of Rebecca's work to create models for data that's not normally distributed and use the model to help predict what other variables will be," he said. 

This predictive power could have applications across a wide variety of subjects, from ecology and economics to physics and quantum science. 

Noah Peterson, 1st year PhD student

Noah Peterson, a first-year PhD student, is working on data assimilation for space weather. 

"We don't often know if we have either all the physics included or if our initial parameters are entirely correct, so sometimes the model is off, and we also have observations, but sometimes those observations can have somewhat large amounts of error in them," he said. 

Data assimilation is the practice of combining those models and observations with different levels of uncertainty to get a more accurate estimate of true conditions. 

Peterson said that in order to get the most out of his PhD, he had to learn to let go. 

"I was so used to putting all of my time into getting perfect grades, but now that's not the most important thing, the research is," he said. 

Rach Washington, 1st year MSCS Coursera 

Rach Washington is a first-year MSCS Coursera student working with Morrison's group, with Bandy as her mentor. 

"In the project I am currently working on," Washington said, "there is a computationally expensive model that works well to describe the system and a reduced model that poorly describes the system. Our goal is to use a less computationally expensive model that describes the system as well as the complex model." 

Washington recommends that students interested in research reach out and try to build relationships. 

"Even as a Coursera student, I am treated as an on-campus student. I would recommend going to office hours. Course facilitators are always kind and happy to help," she said. 

Sienna Amorese, fourth year undergraduate student

Sienna Amorese is a fourth-year undergraduate student studying statistics and data science. Morrison is Amorese's research advisor, while Price-Broncucia has been mentoring her.

"My experience has been wonderful. All members of the group are open to help guide me. It's amazing to see my studies in the real world and set goals that encourage me to explore," Amorese said. 

Amorese is working with Price-Broncucia to create graphical models from NCAR's weather dataset.

"My research focuses on decoupling random climate variables from a weather dataset at NCAR. I've explored the normality and covariance of variables, and am working on creating graphical models to represent their relationships based on their precision matrix," she said.