In our first installment of our Where Are They Now? series, KI checks in with Rice University’s Chris Tunnell four months after May 2019’s Harnessing the Data Revolution NSF Ideas Lab.
Scientists think that about 25 percent of our universe is made up of stuff you can’t see or feel. They’ve named this non-luminescent mystery material dark matter. Some think dark matter is hot (HDM) while others think its cold (CDM). And, yes, some say – you guessed it – warm (WDM). The cold theory has the most support to date. But, even within camp CDM, physicists don’t agree on what it’s made of. It could be made of WIMPs (weakly interacting massive particles). Or, appropriately enough, MACHOs. (We’re not making this up! See for yourself.)
But, to start with, how do you prove the existence of dark matter? If you’re Rice University’s Chris Tunnell, Ph.D., you take a ton of trapped rare, liquified gas and wrap it in a three-story water tank beneath an Italian mountain, of course.
That’s exactly what he’s doing with his xenon dark matter detector (XENON1T). And, thanks to a new collaboration with colleagues he met the National Science Foundation’s Harnessing the Data Revolution Ideas Lab (HDR) in May 2019, he may be able to use machine learning to measure the infinitesimally small perturbations within the dark matter detector. Doing that would not only increase the likelihood of finding evidence of dark matter if it exists, but would help to push the field of data science forward — impacting many other fields.
Big Data Dilemma
Machine learning (ML) is transforming scientific research, helping scientists do everything from search the human genome for causes of autism to create new chemical compounds. But, according to Tunnell, an astroparticle physicist, ML is still underutilized in his field.
“The impact of ML techniques within the physical sciences has been more limited than you would first expect based on how much press and excitement surrounds it,” Tunnell says. For him, the question is: ‘Why hasn’t ML had a bigger impact on the physical sciences?’
Data-Driven Discovery
This was also the question debated by Tunnell and 124 other researchers from a broad range of disciplines who applied and were invited to attend the HDR Ideas Lab. (An Ideas Lab normally includes 25-30 participants. This event was actually four concurrent workshops, each using the Ideas Lab format.) The purpose of the KI-facilitated event was to create interdisciplinary teams capable of developing “new models of data-driven discovery that will allow fundamental questions to be asked and answered at the frontiers of science and engineering.”
At the Ideas Lab, Tunnell met Waheed Bajwa, an associate professor of electrical and computer engineering at Rutgers University, and Hagit Shatkay, a professor of computer and information sciences at the University of Delaware. The three put together a proposal and presented it along with others on the last day. One week later, the new team was one of those chosen to submit an application for funding. And, four months after meeting for the first time, the trio was awarded a $1 million NSF grant that will allow them to take a deep dive into data collected in the search for dark matter — a project that may just transform data science, as well.
Tunnell says attendees agreed that one of the major hurdles was the kind of data generated by detectors such as used in particle physics. For example, the XENON1T is the world’s largest, most sensitive detector of WIMPs, the hypothetical particles believed to constitute dark matter. The detector contains one ton of liquid xenon and is housed deep below a mountain in at the National Institute for Nuclear Physics – Gran Sasso National Laboratory, Assergi, Italy. In terms of radioactivity, Tunnell says, “this is the quietest place in the Universe.”
Detecting Dark Matter
Xenon is quite pure on its own, Tunnell explains. He describes the detector as a ‘bucket of xenon’ about 3 meters high and surrounded by a water tank that stops neutrons and other types of particles from getting in. In addition, being underneath the Italian mountain prevents the penetration of cosmic rays. “So, the whole idea is that you’ve protected the detector enough so that the only thing that can get to your detectors is dark matter.”
The problem is that analyzing the data collected by the dark matter detector presents unique mathematical challenges. If this project is successful, Tunnell says, it will greatly increase the sensitivity to dark matter. “We hope to solve certain challenges with our data analyses that current data science methods do not help us solve.” The team is taking three mathematical approaches to solving the problem posed by the detector’s data.
- Graph Neural Networks (GNNs) – Unlike digital images which are made up of pixels in a lattice-like configuration, the sensors Tunnell uses are distributed in more complicated geometries. GNNs retain a state that can represent information from its neighborhood.
- Bayesian Networks – Bayesian networks aim to model conditional dependence, and therefore causation.
- Inverse Problems – This method allows scientists to infer the physical phenomenon based on data captured by an array of sensors.
He also says he hopes the team can show that his detector is a leader in the search for a specific kind of radioactive decay, called neutrinoless-double-beta decay. “Normally, people have to build dedicated detectors for this measurement. With smarter machine learning, maybe we can use generalizable detectors for multiple fundamental physics measurements.”
Finding solutions to his data analysis challenges would also benefit other areas of research, Tunnell says. “We are trying to make machine learning algorithms for a class of problems that a lot of physical sciences struggle with. So, I think the biggest impact will be beyond particle physics.”
And, to ensure the impact even goes beyond the physical sciences, Tunnell and his colleagues will be partnering with collaborators in fields as disparate as oceanography, neurology and materials. Tapping into these fields means the project will become even more interdisciplinary than it already is, Tunnell says. “If you want to expand out and make this thing five times bigger, then that’s five times as much communication with new people, which is exciting and will most likely also be exhausting.”
Exhausting, yes. But, likely five times as innovative in one-fifth the time. Dark matter, we’ve got our xenon eye on you!