Since completing her PhD in high-energy physics, Jane Cummings has learned more than she ever imagined she would about dairy farming. “I even read dairy journals,” she says.
The switch from proton collisions to cows came in 2017 when she joined Cainthus, a start-up in Dublin, Ireland. The company, which plans to launch its first product later this year, combines two branches of computer science—artificial intelligence and computer vision—to help large-scale farms manage cows’ health, comfort, breeding, and milk production. Cummings is the company’s vice president for computer vision and data science.
Cummings grew up in Brooklyn and went to Grinnell College in Iowa, where she pursued a double major in physics and math, and then to Yale University for a PhD in physics. Many of the skills she developed and techniques she learned qualify as data science, she says. After spending five years at CERN in Switzerland, she decided to stay in Europe and look for a job in industry.
PT: How did you choose your research area in graduate school?
CUMMINGS: In my application to Yale I expressed interest in high-energy physics. It was a shot in the dark. A faculty member, Sarah Demers, contacted me before I started and offered me an opportunity to do some research in her group. She was just starting there herself, and she was looking to get a group of people interested in her research. Coming from a small liberal arts school, I thought it would be good to get some experience in research as soon as possible.
PT: Did you ever waver from your “shot in the dark” path to high-energy physics?
CUMMINGS: I never considered doing anything else. I spent two years at Yale, and in the summer of 2011 I moved to CERN, where I was based for the rest of my graduate school time. I worked on the ATLAS project at the Large Hadron Collider. I really enjoyed working at CERN. It was a culture shock when I got there—coming from a small school and always working in small groups, and then my first real research experience was in this collaboration with thousands of people. It was a big change, but I found it really engaging on many levels.
PT: You must have been there when the Higgs boson was discovered. What was that like?
CUMMINGS: It was really exciting. On the day of the announcement, I was in one of the satellite rooms because I didn’t camp out overnight to get a spot in the auditorium. The energy was palpable. I feel really lucky because I was there basically when the collider was turned on, when we started to have the first collisions, through the Higgs discovery. It was a great time to be based at CERN.
PT: What was your research at ATLAS?
CUMMINGS: I was on a team that was developing the first measurements of tau polarization in a hadron collider environment. We made the first tau polarization measurement in W-boson decays to a tau and neutrino and later measured the tau polarization in Z-boson decays to tau pairs.
PT: Did you consider an academic career?
CUMMINGS: I was not motivated to stay in that world. There is a huge amount of competition for attractive postdoctoral roles, and then it gets even worse for professor positions. I saw a lot of people a few years older than me going through that process, and I didn’t see myself going through that. I was interested in trying new challenges and getting different experiences.
PT: How did your colleagues view your seeking a nonacademic job?
CUMMINGS: My adviser was fully supportive of me. I never felt pressure to stay in academia. But a lot of academics don’t have awareness about outside opportunities. I did a lot of research on my own. I started by reaching out to people I knew from CERN who had already left and gone into industry. I asked about their experiences and how they liked their jobs. That was really helpful.
PT: What’s an example of something helpful you learned by talking to people in industry?
CUMMINGS: I got a lot of good advice around how to best communicate how my experience, background, and skills, and specifically my experience at CERN, could be relevant—so that companies would want to hire someone who knows how to measure tau polarization at ATLAS, even though those words are completely foreign to most people. It was useful to get perspectives on the context outside of academia, to learn what would resonate in both a CV and an interview.
PT: So how did you find Cainthus?
CUMMINGS: Recruiters found me and contacted me through LinkedIn; I had indicated that I was actively looking for a role in Dublin. They gave me a bit of information, and it sounded intriguing. It was quite different from the rest of the roles that I was applying for, mostly at large, multinational tech companies based in Dublin. Cainthus was a small start-up.
I met with a founder of the company the next day, and we ended up sitting over coffee for three hours. We spent a lot of time talking about dairy cows during my interview, which was unlike any other interview I had experienced. I decided to take the job because of the uniqueness of the opportunity and because, despite the ambiguity of the role, I thought I’d really be able to do something quite interesting and challenging.
PT: What does the company do?
CUMMINGS: We are building a computer vision–enabled product that can be installed on dairy farms. We use computer vision and deep learning to monitor animal behavior as well as operations on the farms.
We are introducing cameras and the opportunity to make observations 24 hours a day, seven days a week, and to build intelligence into that monitoring, so it’s as if you had a human standing there who is watching what is happening. You can do things like set an alert for certain scenarios that you wouldn’t want to happen on the farm, and then we can do trend analyses to understand where there is room for improvement.
The farmers we are looking to work with have huge farms with, say, 6000 animals. The idea is to offer a capability to observe what’s happening on the farm at a granular level, which is not possible on farms of this scale without cameras. This is mostly for large dairy farms in North America.
PT: Do you track individual cows?
CUMMINGS: Yes. The idea is that we can look for signals or patterns in individual animal behavior. If an animal is sick, the system could flag her to be checked by a vet or flag her for a particular illness. We are not at the stage where we can do that, but that is the vision. We can also look at the amount of time that each individual animal spends feeding, drinking, lying down, or standing in the lane. This can build up a profile of each individual animal so that the farmer can make use of that information to manage the entire herd.
Farmers already have a lot of herd software platforms to profile their cows and to keep track of different health issues—vaccinations, other treatments, milk production, breeding. They have a huge amount of automation and technology on the farm, and that comes with a huge amount of data, so they are really comfortable working with data that tells them about the farm. Our system will complement such platforms by giving additional information on each individual animal.
PT: Why North America?
CUMMINGS: The local farms here in Ireland are significantly different from North American dairy farms. Irish farms usually have no more than 300 or 400 animals, and there is a lot of pasture-based farming—the cows are indoors in the winter, but they go out to pasture in the spring. We can’t build a product today for those types of farms because we don’t have a way to get cameras covering the pasture. In North America the cows are mostly inside.
PT: What is your role with Cainthus?
CUMMINGS: When I joined in 2017, the company was small and I did a bit of everything. We were doing computer-vision development, prototyping, and testing on a small number of dairy farms. Last year we hired more people and transitioned from an R&D-focused company to a product-focused company. A lot of that transition was marked by completing the proof of the technology, making the business case—that is, showing there was a demand in the market for a solution like ours—and demonstrating that it was technically feasible. I became head of the computer vision and data science team. Back then it was a team of two; now we are a team of six. We are embedded in the products team, which now has a head count of 20. We have grown and changed a lot in the past year.
PT: How do you use your physics background in your current job?
CUMMINGS: I don’t get to apply my specific physics background very much these days, but I use similar research methodologies and my programming background. The ATLAS detector is a large sensor similar to a camera, and so I had a lot of experience working with large volumes of unstructured data through my time at CERN.
PT: Does your experience at CERN and on the ATLAS team help in the management and teamwork aspects of your current job?
CUMMINGS: Absolutely. I worked on many large teams at CERN. A lot of them were distributed, so we had people at different institutions collaborating. I definitely benefited from the experience in terms of being able to run teams today.
PT: Have there been any surprises?
CUMMINGS: Every single day. I think that has to do with working in a volatile, high-growth start-up environment. Things change constantly. Learning how to navigate has been a big part of the experience.
Let’s block ads! (Why?)