Timothy R. Brick, PhD

Research Projects

Technology, Responsiveness, and Real-Time Modeling

Wearables and smartphones are increasingly everywhere. These tools are of course used for all of our activities of everyday life. As a result, they give us an interesting window into the processes at play in our lives. And they can be used to help us make positive changes in our lives. The most visible areas for this are fitness and sleep, where people's watches warn them if they need to walk or track when their sleep hygeine is reduced.

The Wear-IT project, a framework for mobile data collection and adaptation. One primary Wear-IT project is in collaboration with Dr. Hobart "Bo" Cleveland's THRIVE lab. We're interested in addiction recovery--that is, the time after a person breaks free of a substance use disorder and works to thrive without recourse to substance use. The popular model of drug addiction treatment expects people to check into a residential treatment center for 30 days and then come out clean, but the reality is much tougher. In practice, recovery is a life-long process that often requires reinventing one's identity and facing immense challenges. Our work looks at the day-to-day challenges that people in recovery face, including the person-specific emotions, people, and places that trigger craving, and the abilities, grit, faith, and determination that help people push through.

Another project, just wrapping up, is the FamBest project in collaboration with Sunny Bai at the Ballmer Institute. That project takes advantage of the adaptive capabilities of Wear-IT to deliver smartphone surveys right when they're useful to study the interactions between parents and adolescents. For example, after the parent and child spent time together and then separated for a few minutes, Wear-IT triggered surveys to ask about the conversation. Also, when parents and kids got into their cars together, the system nudged them to turn on a dashcam in their car and let us record the conversations they had there. These data get us a beautiful slice of the relationships between parents and their kids, to help us understand the influences that can help promote kids' health and reduce their risks.

Data, Analysis, and Privacy

With a background in computing and a framework dedicated to learning about the details of your life and activities, I'm very interested in data and I'm a bit paranoid about privacy. As a scientist, though, I ask people to trust me with data about them all the time. It's a lot of responsibility. With all this data from your wearable, smartphone, and dashcam, we can learn a whole lot about an individual.

Modern scientific practice requires than we share data with other scientists. But privacy concerns mean we need to keep this invasive data to ourselves, and not share it around. Current analytic practice means that the data are collected in one place, and we deal with the privacy concerns by not letting the data leave that place. But that's bad for science. Pitting privacy and scientific concerns against each other might be the wrong way to go about this.

The MIDDLE project is a proposal that hopes to stop making science and privacy opposed to each other. What if instead of collecting data, we instead left our measurements in the care of the person we measured? Then they'd have access to their own data at all times. The person who owned the data would be the person described in the data. So now your privacy is up to you. All we need is a privacy-preserving analysis method, and we'd be good to go. MID/DLE is the first step towards that method.

Data Mining, Simulation, and Statistical methods for behavioral science data

In order to analyze a lot of the data we get from things like video data and wearables, we need to be able to turn the stream of data we get into understandable information. This requires a wide range of knowhow from the computer science literature, like computer vision models and data mining techniques, but also conventional behavioral science statistics. It also needs a good deal of substantive expertise from the human behavior side of things. And a lot of timeseries modeling techniques from engineering control theory and the related fields.

I tend to do most of my conventional statistics in the Structural Equation Modeling framework, which is part of why I helped develop the OpenMx statistical software. OpenMx (now with over 1.6M downloads) is a toolbox for rapid development of new methods. We've built up tools for it to integrate machine learning approaches like regularization, products of latent and observed variables, tools for easier model understanding, and complex, cross-classified multi-level SEMs.

Neuroimaging and physiological monitoring to understand underlying processes

Neuroimaging tools have come a long way. We're still in the early days of understanding how these tools work and how to use them to really understand what's happening in our nervous system, but there's a lot we can do with the tools that we do have. Most of my interests here are in understanding the moment-to-moment processes at play in cognition, to help motivate an understanding of how it influences behavior. In tandem with neat mobile tools like functional Near Infrared Spectroscopy (fNIRS), other tools like transcranial Direct Current Stimulation (tDCS) (e.g. this one) give us an interesting way to "nudge" the system and try to understand how it works.

But neuroimaging isn't the only way to look at physiology. Peripheral nervous management, via tools like sensors for skin conductivity, temperature, and heart rate on typical wearable devices can help us to understand the give-and-take of emotional processes (like affective arousal) in day to day life.

Generative AI and Analytical Deep Learning

You basically can't get away from generative AI tools and deep learning methods these days. So, naturally, I use a few of them. The generative side of these are mostly helpful for doing just that--generating things. Usually, we leverage things like StyleGANs--those things that let you turn your photo into an oil painting--to build carefully controlled and crossed stimuli [publication forthcoming] to understand how people interpret things like lighting, greenery, and buildings.

But there's an analytical side to deep learning tools as well. Ongoing work targets the representations that deep learning tools--the representations that underlie Large language models like ChatGPT--can be used to help us advance the science of behavior past the world of linear modeling and into something bigger. This project is still new, but I'll try to keep things updated as we make progress on it. C

I also have some other research interests, and some older projects.
And some random academic nerdery.