Q&A with 2022 Meta AI Residents: Research topics, future plans, and more

The Artificial Intelligence (AI) Residency Program is a one-year research training position designed to give talented professionals hands-on experience in AI research while working in Meta AI. This program is ideal for those interested in applying to PhD programs, publishing research papers, and maximizing their experience in AI research as they advance their careers. The AI Residency provides the perfect opportunity for participants, typically recent graduates with bachelor’s or master’s degrees, to experience the day-to-day work of a researcher. Residents can work with Meta AI researchers in areas such as AI-powered translation, diffusion models for video generation, decoding speech from brain activity, and much more.

In this program, participants will be paired with AI researchers to help guide their projects and research problems. The research will be communicated to the academic community through collaboration across Meta AI, academic papers and conferences (e.g., NeurIPS, ICML, ICLR, CVPR), open source code releases, or applications at Meta.

We encourage applications from people with a technical background and who hope to apply to a graduate program but want more preparation before doing so. Prior experience in machine learning is certainly a strength, but we seek people who have a passion for AI and come from a diversity of backgrounds, including math, physics, finance, economics, linguistics, computational social science, neuroscience, and bioinformatics. This is a full-time program that cannot be undertaken in conjunction with university study or a full-time job.

We asked some of the members of this year’s cohort of AI Residents to describe their experiences in the program. (You can also read brief interviews with several Meta AI Residency Program Mentors.)

Abdalgader Abubaker, London AI Resident 2021-2022

“I am working on integrity challenges. The exciting part of the research is its potential impact on Meta social networks to enhance the existing models that are used to prevent integrity violations on the different Meta platforms.”

Alexander Gurung, New York AI Resident 2021-2022

“The program taught me how to drive my own research, and how to come up with new research questions and areas for exploration based on my own intuition. It has also taught me how to tackle larger-scale problems, from the ideation stage all the way through to paper writing and presentation.”

Badr Alkhamissi, Seattle AI Resident 2021-2022

“I am building new language model architectures to enable AI to perform different reasoning skills. Since reasoning is one of the pillars for language models to be deployed in the real world, this research addresses an essential problem for using them in a reliable and responsible fashion.”

Christina Du, London AI Resident 2021-2022

“The program helped me become a stronger researcher. I learned how to ask good research questions and how to design experiments to test our hypotheses. I also got more hands-on experience in writing academic papers and presenting my work to other people.”

Isabelle Hu, Menlo Park AI Resident 2021-2022

“The program gave me the opportunity to work on exciting research directions. As part of FAIR’s multimodal generation research project, I’ve worked on data curation modeling experimentation for text-to-video generation and image-to-image translation. This research area will enhance how people express their creativity in a very fun and meaningful way.”

Ishita Mediratta, London AI Resident 2021-2022

“The program has been a great learning experience for me. I learned how to formulate ideas, present my work to others, and write scalable code. I will apply to PhD programs this fall and hope to continue my collaboration with my Meta Mentor throughout my PhD.”

Jude Fernandes, New York AI Resident 2021-2022

“My biggest struggle with research before the AI Residency was feeling lost with open-ended questions. I’ve been lucky to have Mentors and collaborators who have given me tools that help guide me in these situations. The program has also given me the chance to fully immerse myself in topics I care about, which I have found is the best way to develop intuition and insight.”

Karmesh Yadav, Menlo Park AI Resident 2021–2022

“I am currently working on learning visual representations for embodied navigation and rearrangement tasks using self-supervised learning. The AI Residency helped me hone my technical and interpersonal skills by learning how to orchestrate large-scale experimentation and distributed training, how to collaborate in large teams while still leading and taking ownership of a project, and how to manage contributions over multiple projects.”

Keren Fuentes, New York AI Resident 2021–2022

“I am currently working on quantifying the limitations of widely used fairness metrics that aim to measure bias in masked language models. I find this area of research exciting because it can have a positive impact on people in underrepresented groups.”

Millicent Li, Seattle AI Resident 2021–2022

“My research has encapsulated a wide variety of topics in the natural language processing (NLP) space. During my time here, I’ve been able to meet so many researchers of different backgrounds, take part in a number of projects, and ask many questions.”

Nima Shoghi, Menlo Park AI Resident 2021–2022

“I am on the Open Catalyst Project, which is a collaborative research effort between FAIR and Carnegie Mellon University’s Department of Chemical Engineering. My research involves the development of new machine learning models for molecular property prediction tasks. As someone who does not have a background in chemical engineering, I found things quite challenging at first. However, having the opportunity to conduct research in a challenging new domain alongside a diverse set of highly qualified researchers has proved to be extremely exciting and rewarding.”

Suvir Mirchandani, Menlo Park AI Resident 2021–2022

“My research has focused on vision-and-language pretraining and self-supervised learning, specifically interactive image retrieval: Given an image and a piece of text feedback from a user, retrieve an image that incorporates the user’s feedback. It has been inspiring to be surrounded by so many talented researchers at Meta AI. After the AI Residency, I will be starting my PhD at Stanford University."

Vidhi Jain, Menlo Park AI Resident 2021–2022

“I am focused on building learnable policies for complex, long-horizon tasks in unknown indoor environments. Solving a real-world problem is hard and messy, yet so important, and building solutions geared for it is what makes me excited about this research.”

Zeyu Liu, Menlo Park AI Resident 2021–2022

“The AI Residency program gave me a chance to work with great researchers and receive different perspectives. Additionally, being at Meta AI and FAIR has provided me with the resource to run experiments that are of strong interest to me. After the AI Residency program, I will complete my master’s program and then hope to apply to PhD programs.”

Tackling high-impact research challenges

Abdalgader Abubaker is working with Meta AI researchers in London on integrity challenges. His AI Residency is focused on developing a new approach for graph neural networks to leverage the hypergraph structure — a generalization of graphs with high-level representation in which the edges can connect more than two nodes. “The exciting part of the research is its potential impact on Meta social networks to enhance the existing models that are used to prevent integrity violations on the Meta different platforms,” Abdalgader says. He plans to apply to PhD programs after completing the Residency program.

Badr Alkhamissi is based in Meta’s Seattle office, where he works on responsible AI research. He is building new language model architectures to enable AI to perform different reasoning skills, such as chaining, composition, coreference, arithmetic, and temporal resolution. “Since reasoning is one of the pillars for language models to be deployed in the real world, this research addresses an essential problem for using them in a reliable and responsible fashion.” Badr plans to apply to PhD programs after his AI Residency.

Jude Fernandes is with Meta AI NLP researchers in FAIR’s New York office. Jude has worked with his Mentors to improve the way dialogue models respond to toxic content using controllable generations and has worked on improving fairness in language models by training them on demographically perturbed corpora. “As AI becomes ubiquitous in our world, I feel an urgency to make sure that these systems won’t cause harm to people, and I feel excited about the potential of building a future that isn’t defined by the power structures that exist in society today,” Jude says.

Keren Fuentes also works at FAIR’s New York office, on the Responsible AI team. Keren’s interests include understanding and reducing social biases in language models. She is working on quantifying the limitations of widely used fairness metrics that aim to measure bias in masked language models. “I find this area of research exciting because it can have a positive impact on people in underrepresented groups,” she says. After the program, Keren will apply to PhD programs.

Building research skills

Alexander Gurung is working with Meta AI researchers in FAIR’s New York office toward building text-adventure game language models that can both predict what will happen after a player takes an action (requiring an understanding of the state of the world and how it changes) and produce interesting narrations of the event. They hope this work will one day help train language models that have a grounded understanding of how our world works.

“The Residency has helped me learn to drive my own research, coming up with new research questions and areas for exploration based on my own intuition,” he says. “It has also taught me how to tackle larger-scale problems, from the ideation stage all the way through to paper writing and presentation.” Alexander will complete his last year of his master’s degree after the program.

Christina Du is working with Meta AI NLP researchers in FAIR’s London office. She is exploring new autoregressive solutions to extract a set of entities from an input context. “This project is very exciting since our results can bring potential impact to the NLP community and inspire further research on mention-free entity linking,” Christina says.

Christina says the AI Residency program has helped her become a stronger researcher. “I learned how to ask good research questions and how to design experiments to test our hypotheses from my Mentors. I also got more hands-on experience in writing academic papers and presenting my work to other people,” she says. She will pursue her PhD at the University of Edinburgh in the fall.

Karmesh Yadav works on the Habitat project in FAIR’s Menlo Park office. “The AI residency has helped me hone my technical and interpersonal skills as a researcher. I have learned how to orchestrate large-scale experimentation and gotten well versed with distributed training, how to collaborate in large teams while still leading and taking ownership of a project, how to build on top of other people's efforts and familiarizing them with my own work, and, lastly, how to manage contributions over multiple projects while not getting pulled in every direction,” he says. After the AI Residency, Karmesh will start his PhD in Embodied AI at Georgia Tech.

Suvir Mirchandani works on AI Commerce Multimodal in FAIR’s Menlo Park office. “Through the AI Residency, I have developed skills in writing and structuring research code, sharing research ideas with collaborators, and managing experiments. It has been inspiring to be surrounded by so many talented researchers at Meta,” he says. After the program, Suvir will be starting his PhD at Stanford University.

From predicting molecular properties to teaching robots to load dishwashers

Vidhi Jain joined Meta AI researchers working on problems in robotics and embodied AI. Her AI Residency has focused on building learnable policies for complex, long-horizon tasks in unknown indoor environments. “My focus is currently on dishwasher loading based on different human preferences. Solving a real-world problem is hard and messy yet so important; and building solutions geared for it is what makes me excited about this research,” she says. Jain plans to pursue her PhD at Carnegie Mellon University once she’s completed her AI Residency.

Isabelle Hu is part of FAIR’s Make-A-Scene research project. “I’ve worked on modeling experimentation for text-to-video generation and image-to-image translation, and recently I’m working on curating a large scale video-text dataset using multimodal models. This research area connects the domains of vision and language, and it will also enhance how people express their creativity in a very fun and meaningful way.”

“The residency program gave me the opportunity to work on exciting research directions that I had little previous experience in and to experience doing AI research at a much larger scale than in academia,” she says. Hu will continue working in the industry and join a biotech company as a machine learning scientist once she’s completed the program.

Ishita Mediratta researches reinforcement learning (RL) at FAIR London. Taking inspiration from knowledge distillation, she is working to improve systematic generalization of RL agents, such as leveraging Unsupervised Environment Design for automatic curriculum development and open-ended learning. She’s also using Procedurally Generated (PCG) environments to test generalization capabilities of RL agents.

“The residency program has been a great learning experience for me,” she says. “I got to learn how to formulate ideas, present my work to others, and write scalable code.” Mediratta will apply to PhD programs this fall and hope to continue her collaboration with her Mentor through the program.

Zeyu Liu focuses on NLP at FAIR’s Menlo Park office. Zeyu’s recent project developments include “scaling up” with “mixture of experts” techniques. Zeyu and his team proposed a unified framework to characterize different exploration in this region with two variables. “We show that the upper bound of the transformer is far from being reached. And we further study the impact of the two variables to inform future architecture exploration,” says Zeyu.

Millicent Li is on the NLP team at FAIR Seattle, and her research has encapsulated a wide variety of topics in the NLP space. “I’ve been able to meet so many researchers of different backgrounds, take part in a number of projects, and ask questions based on what I’m interested in learning about. The program has let me let my curiosity drive the work that I do without strings attached,” says Millicent. After the program, Millicent will be starting her PhD in computer science at Northeastern University working at the intersection of NLP, human-computer interaction, and healthcare.

Nima Shoghi is on the Open Catalyst Project at FAIR MPK. Open Catalyst is a collaborative research effort between FAIR and Carnegie Mellon University’s Department of Chemical Engineering. “My research involves the development of new machine learning models for molecular property prediction tasks. As someone who does not have a background in chemical engineering, I found things quite challenging at first. Once I got past the onboarding stage, however, having the opportunity to conduct research in a challenging new domain alongside a diverse set of highly qualified researchers with backgrounds in machine learning and computational chemistry has proved to be extremely exciting and rewarding,” says Nima.

Click here to apply for the 2023 cohort.