The Future of Science is Here
When Rebecca W. Doerge, Glen de Vries Dean of the Mellon College of Science, talks about the future of science, she likes to tell a story about her Grandma Ruth. Ruth was a telephone switchboard operator who manually connected people using telephone cords. When the rotary phone was invented, that technology “cut the cord” making the traditional switchboard obsolete. After the cord was cut, telephone technology advanced quickly to where most of us now carry a phone that is basically a small computer with us at all times.
When she thinks about the future of science, and how MCS should prepare for the future, Doerge asks: What cords can we cut to advance science? How can we cut them?
She believes that the answers to these questions lie in automation, computation and artificial intelligence. These three areas also happen to be ones where Carnegie Mellon and MCS are uniquely poised to make a dramatic impact by applying the university’s strengths in computation, engineering and data science to the foundational sciences. The result, says Doerge, will change the way that science is done and advance discovery.
Examples of this eye to the future can already be seen throughout MCS, with researchers using automation, computation and artificial intelligence to advance their fields.
Science is undoubtedly moving forward at an ever-increasing pace. A pace that is so fast that humans can’t keep up without assistance. The answer to keeping the momentum of discovery moving forward lies in automation, which will allow researchers to conduct more ambitious experiments, collect massive amounts of highly reproducible, uniform and reliable data, and create more sophisticated models.
Currently, MCS is boosting its capacity for automation by building an automated lab space in the Mellon Institute. The space will be used by a number of researchers that span disciplines, including Chemistry Professor Stefan Bernhard.
“Chemical research is becoming so complex that you need to do many experiments to explore a given parameter space. This is where automation comes into play,” said Bernhard.
The automated lab space will use robots to dispense chemicals, conduct numerous reactions at the same time and analyze those reactions. For example, Bernhard develops catalysts that can photochemically split water molecules to liberate hydrogen gas, which can be used as a solar fuel. Using automation, he and his colleagues can run exponentially more experiments and then use machine vision, machine learning and other big data tools to sift through this data.
The automated labs will bring together researchers in the sciences with those in other disciplines including machine learning and engineering.
“The automated lab is a nexus for collaborative efforts by connecting people with automation,” said Bernhard.
Algorithms for Life
From the cellular circuitry in our brain to the rules that underlie evolution, there are algorithms found in nature that can have valuable applications for technology.
For example, Biological Sciences Professor Alison Barth is teasing apart the algorithm by which cortical circuits receive sensory information and how they adapt to it to learn.
“The brain is receiving information all the time, and it transforms that information to drive some type of motor output,” said Barth. “We’re thinking of the brain as a computational device that can change its computations based on experience. We want to know what is the brain’s algorithm for learning.”
Finding out that algorithm can help researchers who are creating engineered systems, especially those that employ deep learning.
Barbara Shinn-Cunningham, director of the Carnegie Mellon Neuroscience Institute, works to identify the algorithms of auditory signal processing, with the goal of creating a better hearing aid that can turn down some sounds, like ambient room noise, while turning up others.
The end result of both researchers’ work, as well as the many other faculty members who are studying the algorithms of life, will be new technologies that capitalize on what has already been perfected by years of evolution.
MCS is also making important contributions to the rapidly growing field of artificial intelligence. Many of the automated machine learning algorithms used by computer scientists are poorly understood. The Mathematical Sciences Department, through its interdisciplinary program in algorithms, combinatorics and optimization, is studying the broad foundations of data science to improve our understanding of the algorithms that fuel artificial intelligence.
“As we become more reliant on algorithms to inform decision making and make predictions in fields like health care, infrastructure and business, we need to not just accept that they work. We need to know how they work,” said Tom Bohman, Alexander M. Knaster Professor and Head of the Department of Mathematical Sciences.
Active Machine Learning
Almost every area of science has benefitted from better data collection technologies, whether it be high throughput screening and sequencing technologies or higher resolution microscopes and telescopes. These technologies result in increasingly larger data sets, which can be daunting — if not impossible to sort through.
“Machine learning is poised to have a significant impact in research across science and engineering, changing both the types of questions that can be asked as well as how they are answered,” said Associate Professor of Chemistry Newell Washburn, who organized the Machine Learning in Science and Engineering conference held at Carnegie Mellon last year.
Carnegie Mellon researchers, with their world-renowned expertise in computer science, machine learning, and statistics and data science, have quickly established themselves as leaders in making sense of these complex datasets, by applying machine learning to data from a wide variety of fields.
In his research, Washburn has created machine learning models that can predict the behavior of complex physical systems and that can be used to speed up the materials design process.
Astrophysicists in the McWilliams Center for Cosmology utilize machine learning to make sense of the complex data coming from some of the world’s largest cosmological surveys. Physics Professor Rachel Mandelbaum uses machine learning to find and interpret evidence of weak gravitational lensing, one of the most direct, but also most difficult, ways to learn about dark matter and dark energy.
“Machine learning is invaluable to helping us pull information out of huge data sets, which can contain tens of millions to eventually billions of galaxies,” said Mandelbaum.
A Revolution in Science
MCS’s progress so far, combined with the university’s unique areas of expertise, position Carnegie Mellon to revolutionize the future of science.
“The future of science lies in combining the foundational sciences with automation, big data and technology,” Doerge says. “No other university is positioned to do this as well as Carnegie Mellon.”
♦ Jocelyn Duffy