October 24, 2022

Q&A with Akshitha Sriraman, Assistant Professor, Department of Electrical and Computer Engineering, Carnegie Mellon University

By: Meta Research

In this monthly interview series, we turn the spotlight on members of the academic community — and the important research they do — as thought partners, collaborators, and independent contributors.

For October, we are spotlighting Akshitha Sriraman, Assistant Professor, Department of Electrical and Computer Engineering at Carnegie Mellon University (CMU). Akshitha is a longtime Meta collaborator, having been awarded the 2020 Meta Distributed Systems Fellowship while working toward her PhD at the University of Michigan. She has coauthored multiple academic papers with Meta researchers, participated in two Meta-sponsored internships, and served as a part-time research scientist at the company. Most recently, Akshitha was among the winners of the 2022 Systems Research request for proposals.

In this Q&A, Akshitha shares how her nontraditional path to research became her superpower and how her passion for equity — in her research, her lab, and on campus — drives her every day.

Q: What inspired you to pursue a career in research?

Akshitha Sriraman: My path to research in computer science was fairly nontraditional. After studying electrical engineering rather than computer science as an undergraduate in a small state school, I realized I needed more hands-on experience. While pursuing my master’s degree at the University of Pennsylvania, I taught myself how to write code through YouTube videos and online courses. Ultimately, I earned a B+ in my first computer architecture/systems course — a class I now teach as a professor!

Eventually, I found a mentor, who was an assistant professor at the time. Collaborating with him got me excited about pursuing a PhD and a research-oriented career.

Q: When were you introduced to Meta?

AS: I met Kim Hazelwood, director of Engineering, AI, at Meta at a conference. I was navigating feelings of imposter syndrome, but Kim somehow thought I had potential and encouraged me to apply for a Meta internship. Since then, my experience at Meta has not only helped me produce my body of research but opened my eyes to the real-life impact of my work. A significant portion of my PhD work happened because of Meta, and Kim has been a strong mentor ever since. If we hadn’t talked at that conference, my whole research career would be very different.

Q: Can you tell us more about your collaboration with Meta over the years?

AS: My first internship at Meta, where I was mentored by Abhishek Dhanotia, redefined how I develop research problems. I produced a top-tier research paper with Meta, profiling production applications and minimizing hardware bottlenecks, and I designed a prototype that teams at Meta went on to develop further. I was then offered another internship with Meta, where I researched how to systematically understand hardware acceleration opportunities at scale. Within three months, we successfully characterized acceleration opportunities, how to systematically model them, and wrote a paper that was recognized as one of the 12 best computer architecture papers that year.

My teammates at Meta taught me the joy of research. For me, this joy came from discovering that a small percentage improvement in data center efficiency can produce a meaningful change in cost and environmental impact.

One of the main reasons I wanted to become a professor was so that I could open doors for students who wouldn’t otherwise have the opportunity. Talent is equally distributed, but opportunities are not.

Q: You’re passionate about hyperscale efficiency — can you tell us about your current research?

AS: Applications are growing at a rapid pace, yet hardware performance isn’t scaling at the same rate. At CMU, we’re focusing on how to efficiently design hardware and software systems to enable the apps of tomorrow, without compromising key metrics like cost, energy, sustainability, and equity. Our goal is to rethink the computing stack across hardware and software systems to achieve these metrics, including saving time and money, reserving more energy, reducing hardware waste, and not discriminating against certain users.

Our research group is the first in the systems community to introduce equity as a first-order metric for systems design. Instead of continuing today’s trend of building web systems for the top 1 percent of the population, we are enabling new equitable solutions by building systems for rural communities, to lift more than a billion people out of poverty and open new economies and markets. In collaboration with CMU-Africa, we are building services for rural populations (e.g., online farmers markets) and redesigning systems to support them (e.g., developing systems that automatically decide where to compute, depending on device/network constraints). We plan to deploy these systems in sub-Saharan Africa, connecting rural populations in more meaningful ways.

Additionally, we are using equity as a design metric to identify and mitigate existing web systems’ inequities. Our work combines bias metrics (e.g., age and race) with user studies to measure performance benefits from biased decisions, and from there building systems that mitigate inequitable responses.

Q: How are you supporting the next generation of researchers at CMU?

AS: One of the main reasons I wanted to become a professor was so that I could open doors for students who wouldn’t otherwise have the opportunity. Talent is equally distributed, but opportunities are not. I also spend time mentoring students from underserved communities. My advice to PhD candidates is to use every single opportunity available, and if there aren’t any, put yourself in positions where you can create opportunities and open doors for yourself. That’s why I’ve created programs that match underserved students with PhD student mentors and disseminate resources. By empowering nontraditional students to pursue their careers and dreams, I hope to help build a more equitable systems field, with these students leading the way.