From the Earth to the Cloud: Felix J. Herrmann has a Vision for Interdisciplinary Research at Georgia Tech

Professor Felix Herrmann is new to Georgia Tech and holds joint appointments in the schools of Computational Science and Engineering, Electrical and Computer Engineering, and Earth and Atmospheric Sciences.

Herrmann comes to Georgia Tech from the Department of Earth, Ocean, and Atmospheric Sciences at the University of British Colombia (UBC) where he served as a professor and the director of the Seismic Laboratory for Imaging and Modeling (SLIM) – a world leader in the development of the next generation of seismic acquisition and imaging technology for the oil and gas industry.

Herrmann was recently invited to be a distinguished lecturer for the Society of Exploration Geophysicists (SEG), which involves a global tour from January to June of 2019. The prestigious lecture series will cover information about the use of compressive sensing exploration seismology.

As Herrmann’s joint appointments stretch across three diverse schools, his research cannot be compartmentalized into one category. In a recent Q&A, Herrmann talks about his research inspirations and what he envisions for the future of these interdisciplinary fields.

 

Q: Exploration seismology, the branch of science concerned with imaging the Earth, is an intriguing field to study. What drew you to this field of research?

A: When I was an undergraduate in engineering physics at the Delft University of Technology, I had to select a topic for my thesis, and I ended up having to choose between image processing and seismic imaging. I choose the latter because I felt that the physics of waves encoded in the wave equation would at least keep me grounded. Subsequently, I got hooked, and since then, I have been amazed by the rich geophysical, mathematical, and computational problems this field has to offer. Seismic imaging really ranks amongst one of the most challenging imaging problems out there, which makes it fun especially since people from industry do care.

 

Q: You were previously the director of SLIM, a laboratory that conducts research in exploration seismology with support from the energy industry. Can you tell us a little about how you advanced into this role?

A: After a short fling working on global seismology at MIT, I returned at UBC and started SLIM. The decision to start a lab channeling industry funding was primarily driven by lack of federal funding in Canada towards larger research programs in the field of earth sciences. The public-private partnership I started helped me to overcome this situation and I am very grateful for the support I have received from the oil and gas industry and matched funding from the Canadian funding agency NSERC over the years. This allowed me to build a group widely considered as one of the world leaders in computational exploration seismology.

 

Q: You are also a founding member of the International Inversion Initiative – an initiative designed to advance the capabilities of 3D Full-Waveform Inversion (FWI) and related technologies. Can you describe what the FWI does and why advancement of its capabilities are needed?

A: FWI is a technique invented back in the early 1980s, which is now causing excitement thanks to recent major improvements in computational power rendering.  Contrary to conventional imaging techniques – mostly based on figuring out how the Earth looks like by considering how long it takes for waves to travel within the Earth subsurface – FWI considers both travel times and amplitudes of transmitted, reflected, diffracted, and refracted waves making it in principle possible to find out what the Earth looks like exactly. In short, FWI produces a digital impression on how the wave speed varies in three dimensions. Despite early successes, FWI is hampered by the fact that it can easily lead to wrong results if our initial inception of what the Earth looks like is wrong. Clearly, this defeats the purpose of FWI and researchers around the world are working hard on building FWI technology that can create reliable images at low computational costs without relying on having detailed information on what to look for beforehand.

 

Q: Now that we have explored into your past, let’s discuss the now: What brought you to Georgia Tech? Are there any current projects that you can tell us about?

A: My decision to come to Georgia Tech was driven by three key factors. First, I want to explore imaging problems other than seismic, such as medical imaging. I know from past attempts towards a crossover that this may be challenging, but I feel that the interdisciplinary nature of research and development at Georgia Tech will be conducive to making this shift. Second, the field of computational seismology is extremely computing-intensive. My lab can basically be seen as a lab where we conduct many computational experiments, each of which requires thousands of CPU hours. The emergence of the cloud and the vision of Georgia Tech to explore this new resource was another key factor in my decision to join the faculty of Georgia Tech. And, finally there is much more willingness at Georgia Tech to explore new models for public-private partnerships to support research and to drive innovations in the computational data sciences whether is directly from the Institute or indirectly via startups.

 

Q: What do you hope to accomplish through your current projects?

A: I think that the cloud will be the great equalizer when data and compute intensive technologies are concerned. In the past, only large companies had access to large data-intensive compute. But now, with the advent of the cloud, everybody does, and this will completely change the game. For this reason, I plan to leverage the cloud to help companies to at-scale validate and apply algorithms being developed at Georgia Tech. The idea really is to monetize certain algorithmic developments by implementing them at scale. In return of these efforts, we expect to get access to compute and funding for our students and scientific software support. I feel that this may be a sustainable new way to fund research and drive innovations while working with industry.

 

Q: What are some of your other research areas of interest?

A: Aside from working on FWI and machine learning related techniques, I have a keen interest in developing new economic and low-environmental impact data acquisition adapting techniques from Compressive Sensing. FWI is only as good as the data you collect in the field. My group is widely considered as the academic leader in this field. More generally, I think I am good at recognizing new developments in the mathematical and computer sciences. I mix and play with these developments to solve applied problems in the imaging sciences. Along the way, I invent new methods that are of interest to these fields closing the loop. To do this successfully you need to have a good understanding of the domain to really know what matters and a willingness to learn to speak the language of both theoreticians and practitioners. That’s where my interests really lie and where I feel I can make key contributions.

 

Q: Did any students or research scientists travel with you from the University of British Columbia or any other previous institutions? If so, who are they? Can you briefly describe why?

A: Yes, Ph.D. students: Mathias Louboutin, Ali Siahkoohi, Shashin Sharan, Mengmeng Yang, and post-doc Rajiv Kumar, have followed me from UBC. They are working on topics ranging from seismic acquisition and imaging, to the development of a domain-specific language for wave simulations, and the application of generative adversarial networks to modeling and acquisition problems. I think they followed me because of the Institute’s unique leading position in the field of machine learning and associated improved prospects on the job market. I also think they like to work with me.