Below is an interview with Li Yin, a Research Scientist at Facebook. Learn about her role, what she does, and in particular how there are different branches of research at Facebook AI.
FAIR (Facebook AI Research) Labs for example are more focused on pure research, whereas Li and her time are doing more applied research (under Facebook AI) in which there are specific use-cases in mind.
The cool thing about working at Facebook AI as an applied researcher is that you not only have a big impact on the research community through the work you do, but you’re also able to have a big impact on Facebook’s product direction and the millions of users on the platform.
Here’s how Li’s experience has been thus far.
I work at Facebook AI, particularly applied research, as a Research Scientist, developing computer vision models for AR use cases. Our models eventually go from research to production.
A lot of our work also revolves around data - so it’s not just modelling. For example, in computer vision, we need to figure out what data we train our models with, what is important for the user, and how to annotate this data (for ex: what’s the timeline). So a lot of this work usually involves collaboration with data scientists and data engineers.
Since our models eventually go into production, we also work closely with product managers to fine tune different use cases and implementation practices.
I’m based in Menlo Park, California, USA.
There are many different roles that Facebook AI hires, such as these categories:
We either directly hire people into our teams, mostly Research Scientists or we hire from boot camps where people are hired into facebook but not positioned at a particular team. At facebook, most are hired without team selection.
Direct hires into teams are likely high stake engineers or researchers with particular research backgrounds that the teams are early looking for. So in this case, the recruiter would reach out to published researchers in the field in which Facebook is looking to hire.
Compared with FAIR labs who work on pure research goals with a ~5 years vision, our project is often < 2 years long, and each project is started with particular use cases in mind such as product surfaces where our models will be launched on, deployed in cloud or on-device. Our work consists of both research and MLOps and our time is split into 3 chunks (1) data collection and data annotation, (2) modeling, and (3) work cross-functionally with product teams to scope proper use cases.
Research that is conducted under FAIR labs will often not even have a use case. But for us, we know that there is a real-world use case that our research needs to satisfy. So for example, within computer vision, our image detection model might have the specific use case of trying to identify fraudulent ads in mind.
It’s cool to get to work both on the research side and the production side. The beginning of a project is obviously more research heavy, and then towards the end we collaborate more and more with other stakeholders.
The contrast between both types of tasks can be challenging at times. This is because when you do research, it’s a different mind state to when you have to actually identify and map out use cases that real world users will be impacted by. The latter is more of a product engineering or product management type task.
So you get to talk to all kinds of different people, like some who are actually creating apps that users interact with, others like UI designers who are crafting the user experience, etc. I find all of this really enjoyable.
One of my projects is related to object detection based on different user personas. This is a novel and challenging research project that doesn’t have too much current literature in the academic space, so it’s exciting.
This is a project that we’ve already secured 2 or 3 different use cases for and we are currently in the research phase.
Each project has different time periods. During research mode, as I mentioned earlier, you don’t really want to do too much cross-functional collaboration because you’re heads down reading academic papers and doing a lot of computational heavy work.
But during the planning stage, so even before the research has begun, there’s a lot of room for discussion with other colleagues and you’re able to bounce ideas off each other. This process can be really fun too.
Finally, it’s important to remember that the research we do is very useful for real-world users, so there’s an opportunity for big impact. There’s also an opportunity to shape the product direction of the company.
At the beginning, I thought that I was just a “PHD” person and that I wanted to stay in academia. Note that I was also an international student and I skipped my Masters degree, going straight from undergrad to a PHD program.
But basically, my advisor at the time and myself didn’t share similar beliefs on how we wanted to conduct our research. I wanted to collaborate more with others whereas my advisor believed in doing work much more independently. And I want to stress that the biggest part that will define your PHD experience is your advisor, so you want to be sure you really gel with them. So ultimately, in my case, I just wasn’t able to get the support and tools I needed to succeed from my advisor.
Then, after doing a few internships in the computer vision field at companies like Nvidia, I realized that I could just learn what I wanted to learn in industry, rather than in academia. So I just dropped out from my PHD program and haven’t looked back since.
Yeah, I’m able to work with and have access to researchers at the top of their fields here at Facebook. I’m able to work with experts in computer vision, data engineering, and really whatever topics that we are exploring. It’s extremely collaborative and that’s what I was missing at my PHD program. Not to mention that at Facebook the work you do is based on real inputs rather than fake academic datasets - you’re actually to build models that solve problems for millions of users and have a real impact on the world.
Even at Facebook, on our research team, we’re not too overly preoccupied if you have a PHD or not. As long as you’ve proven that you can do the job, that’s enough.
Based on the different stage of your project, you’re doing different things. For example, right now, I’m more focused on modelling and I have less meetings than I would if I was in the latter stage of the project. Generally, though, Facebook AI has a very open environment so you’re able to attend as many meetings as you want if you just want to learn from others.
Usually, my day will consist of team meetings, cross-functional meetings, and I will reserve some time for research and modelling. In terms of time-split - it’s usually 30% meetings and 70% modelling / research right now.
We also have one day every week (currently on Wednesdays) where everyone is trying to be more meeting free.
I will say that remote work due to the pandemic has often blurred the line between work and free-time because there’s often no end to research (there’s always something you can do!). And all of us enjoy our work, so it’s very easy to fall into the trap of overworking and burning out.
I think that as an intern, in general, you skew more towards the work side of things because you’re trying to prove yourself and get a return offer. So interns at companies like Facebook and Nvidia often do a lot of overtime.
But overall, it’s important to remember that different people have different working styles. Some people derive a lot of meaning from their work, so they often don’t see the long hours they spend working as a “bad thing”. This is especially the case with smart researchers you work with at Facebook AI - these are individuals that are at the top of their fields and love to learn.
Our managers do often ask us whether we’re ensuring that we have a good work life balance in place, but we never know how to answer that question because we’re so engaged with our work!
My favorite thing has to do with the impact that I can have. Both the impact for the research community and the impact on the product side are huge - they’re equally exciting. I also don’t want to just join a pure research team because I’ve come to enjoy the production side of our projects as well.
Something that’s pleasantly surprised me about working at Facebook has been how open the culture is. Your managers are really invested in your personal growth and want to make sure you find your work meaningful. It’s also really common to discuss career paths with your managers so that you’re always headed in the direction you want.
Personally, I’ve really enjoyed growing in so many ways outside of research (which was what my background mainly consisted of). You have to work with different people, manage conflict, improve your communication skills, and be a team player. These all areas I’ve grown in tremendously.
Some people call it a newsletter - I call it a good time. I write about tech careers and how you can get ahead in yours. It’s my best content (like this case study) delivered to you once a week.