Opportunity as PhD or postdoc on cognitive science and AI at Johns Hopkins University

How do I apply to work with you?

Thank you for your interest in our lab! There are generally two ways to work with us: as a PhD student, or as a postdoctoral researcher.

For prospective PhD students: Please apply for the PhD program in Cognitive Science or in Computer Science at Johns Hopkins University. The FAQs in this document contain more details about the application process.

Applications are submitted through the Johns Hopkins website. You do not apply by emailing me your materials, nor will this help your application. I will review all applications that mention my name as a faculty member of interest.

For prospective postdocs: Please email me directly to inquire about open positions. Please outline your specific research interests and how they would fit in with the goals of the lab. You are also strongly encouraged to identify potential extramural funding sources.

In the future, there may also be opportunities for undergraduate or postbac research assistants. More information will be available soon.

Should I apply for the Cognitive Science or Computer Science PhD program?

Our lab is officially situated within the Department of Cognitive Science, part of the Krieger School of Arts and Sciences at Johns Hopkins. We will have physical lab space in Krieger Hall on the Homewood Campus.

Practically speaking, this means:

• I can be your primary advisor if you are enrolled in the PhD program in Cognitive Science.
• I can be your co-advisor if you are enrolled in the PhD program in Computer Science. You will still need a primary advisor in Computer Science.

You will be able to do research at the intersection of cognitive science and AI, no matter which department you are in. However, your official departmental home does matter in many other aspects, like funding, coursework, and teaching, as well as who your peers and mentors will be. You can find detailed information about each PhD program’s formal requirements online.

The Cognitive Science program might be a better fit if you are interested in human language and cognition, and/or you want to understand AI models from a scientific perspective. The Computer Science program might be a better fit if you are primarily interested in NLP or building AI technologies for practical tasks.

What skills do you look for in a PhD student?

People come to computational cognitive science from many different backgrounds. There is no single profile that makes someone “the right fit”. But in terms of specific skills, our research typically involves the following:

• Programming in Python (and R)
• Deep learning (PyTorch, Huggingface, etc)
• Probability and statistics (data analysis, Bayesian modeling, information theory, etc)
• Human behavioral experiments
  • Web-based experiment development and hosting (jsPsych, Prolific, etc)
  • Principles of experimental design

Many of these skills can be learned through PhD research and coursework, so don’t worry if you only have experience with some of them. I recognize that opportunities and access are not distributed equitably, and every student has a unique background and set of experiences. I take all of this into account when reviewing applications.\

How should I prepare my PhD application materials?

There are many fantastic online resources about this topic, so I won’t repeat what they’ve already said:

• Lucy Lai (neuroscience-focused)
• Mor Harchol-Balter (computer science-focused)
• Sokol-Hessner Lab (psychology-focused)

In addition to the general advice above, I look for the following:

• Can you formulate deep but well-scoped research questions?
• Can you work independently to approach unstructured problems?
  • “Independent” doesn’t mean “alone”. In most cases, we will be working together and with other collaborators. What I want to see is whether you will take initiative to come up with new ideas, iterate on them, and “own” your part of the project.
• Our lab is brand new, and the first few members (like you!) will set the tone for the rest of the lab for years to come. How will you help shape our lab’s culture and environment?

What is your approach to mentoring?

Science does not unfold on a linear path. There may be dead ends, or multiple ways to get at a solution. There will be many different types of challenges (e.g., technical issues with implementations, analyzing experimental data, or paper-writing). My goal is to help unblock you through all steps of the process. Sometimes, especially earlier on in your career, this will look like direct collaboration (e.g., I will help create stimuli, analyze data from a study, or write the Methods section of a paper draft). Sometimes this will look more like top-down guidance (e.g., I will suggest a general idea, and you will come up with concrete ways to implement it). Ultimately, though, I want to “teach you to fish”, helping you gain the tools and skills needed to begin independently pursuing these directions as you progress in your career. I want each of my mentees to find their unique research voice, and take ownership of their own corner of the research world.

I am new to mentorship, and many of you might be new to research. In these early years, we are setting the tone for a brand new lab! This is an exciting opportunity for all of us to learn from each other. There may be times when we are unsure of how to proceed, or when we make mistakes. If at any point something about our mentoring relationship is not working, please do not hesitate to let me know. I promise to listen and to do my best to address it.

What projects will I work on in the lab?

The best way to get a sense of our work is to look at our recent publications. Students are not “assigned” to projects. I work with each student individually to come up with project ideas that serve their interests and long-term goals, while also aligning with the mission of the lab.

Broadly, here are core research areas of our lab:

Cognitive evaluation of AI models

• We use theories, methods, and frameworks from cognitive science to better understand the abilities and limitations of modern AI models.
• In practice, this often means developing benchmarks or evaluation/interpretability methods to measure specific cognitive abilities or computations.
• We also perform targeted manipulations to understand how different kinds of knowledge are acquired. What kinds of input data, training objectives, inductive biases, and cognitive architectures are sufficient or necessary for successful learning? This line of work also connects to cognitive theories.

Computational psycholinguistics

• We use computational tools to study the representations and algorithms that support human language learning, comprehension, and production.
• In practice, this often means running/analyzing data from human experiments, and developing formal models that quantitatively predict humans’ patterns of behavior. We also analyze neural data (such as fMRI/EEG) when relevant, but neuroscience is currently not the primary focus of the lab.
• Specific topics of interest include: pragmatic communication, multimodal learning/communication, social intelligence, resource rationality.

Source and more details: tinyurl.com/bde6syjz