Sensing to Intelligence: Scientific Discovery
By designing autonomous systems that can sense and process data—and, ultimately, design new experiments—Caltech scientists and engineers will spur discovery in areas ranging from chemistry and materials science to biology, space science, and many other fields.
A few projects now under way:
Videos that detect internal properties
Photo: By analyzing small motions in a video, scientists can recover the internal material properties of an object, such as a solid core within an object that appears to be homogeneous. Credit: Berthy Feng/Caltech
Detecting defects inside objects just by filming them
All objects vibrate, even those that appear motionless to the human eye. And because the structure and material properties of an object influence these vibrations, it would be possible to detect interior defects just by looking closely—if only humans could see well enough.
Katherine L. (Katie) Bouman, who specializes in imaging the imperceptible, aims to overcome this challenge by integrating algorithm and sensor design to create exquisitely sensitive imaging systems.
Berthy Feng, a graduate student in Bouman’s laboratory, is developing a set of algorithms that will enable scientists to measure the properties of an object composed of different constituent materials simply by filming it as it vibrates. This technique would replace more expensive scanning laser vibrometers, ultrasonic monitoring, and more destructive methods that require cutting into the object.
Other collaborators in this effort include Caltech materials scientist and mechanical engineer Chiara Daraio and mechanical engineering graduate student Alexander Ogren.
Katherine L. (Katie) Bouman, Rosenberg Scholar and Assistant Professor of Computing and Mathematical Sciences, Electrical Engineering, and Astronomy
Chiara Daraio, Professor of Mechanical Engineering and Applied Physics
Berthy T. Feng, Kortschak Scholar and graduate student, Computing and Mathematical Sciences
Alexander C. (Alex) Ogren, graduate student, Mechanical Engineering
Automated behavioral observation
Photo: A fly responds to visual cues while tethered in a flight simulator. Credit: Floris van Breugel (PhD ’14)
Computers that help researchers investigate animal behavior
Michael Dickinson investigates how fruit flies navigate, control their flight, and act on what they sense. Neurobiologist David Anderson studies fruit flies, too, but his goal is to trace links between genetics, neural circuitry, and emotional behaviors such as aggression.
Both biologists’ work requires time-consuming observations of fly behaviors. So, the researchers have collaborated with computational-vision expert Pietro Perona to automate the process. The resulting vision systems and statistical techniques have made it possible for computers to detect and analyze flies’ videotaped activities two orders of magnitude faster than humans could.
David Anderson, Seymour Benzer Professor of Biology and director of the Tianqiao and Chrissy Chen Institute for Neuroscience at Caltech
Michael Dickinson, Esther M. and Abe M. Zarem Professor of Bioengineering and Aeronautics
Pietro Perona, Allen E. Puckett Professor of Electrical Engineering
The first “radio camera”
Photo: Arrays of inexpensive receivers with powerful correlators have put Caltech’s Owens Valley Radio Observatory at the forefront of discovery. Credit: Chuck Carter
Telescopes that instantaneously follow up on their observations and generate images
At Caltech’s Owens Valley Radio Observatory, astronomers Vikram Ravi and Gregg Hallinan aim to define the future of intelligent radio astronomy. They are creating new arrays that process their own data.
One array compares more than 700 signal paths from 352 antennas to search for radio emissions produced by exoplanet magnetospheres, a potential ingredient for habitability. The other array, soon to grow from 10 to 110 dishes, processes and searches its data for millisecond radio bursts that could yield insights about the evolution of stars and galaxies and the composition of the cosmos.
Traditionally, astronomers have to manipulate data produced by radio telescopes in order to create final images. These new arrays are the testbed for algorithms, developed with support from Schmidt Futures, that integrate signals from many sensors to produce images directly. The goal is to build a 2000-dish array that will be the first true “radio camera.”