michael j tarr

I will be considering new PhD students for fall of 2027 in both Neural Computation & Cognitive Neuroscience

Note that the Neural Computation PhD also offers the possibility of a Joint Neural Computation/Machine Learning PhD - one applies to the joint program after successfully completing several required ML classes with a sufficient GPA

CMU recently hired Maggie Henderson, Jenelle Feather,
Jonathan Tsay, Xaq Pitkow, and Aran Nayebi

You should reach out to them as well.

courses I teach:

Reverse Engineering Brains and Machines 85-426/85-726

Foundations of Brain and Behavior 85-219

interests in natural and artificial systems:

• the inductive biases of biological systems that enable intelligent behavior

• better articulated models of the intermediate and high-level vision

• the functional organizational principles of vision

• the interaction between vision and other modalities

• tools drawn from cognitive science, machine learning, computer vision, computer graphics, and large-scale neuroimaging, including fMRI, DTI, MEG, EEG

Bio. My research spans cognitive science, cognitive neuroscience, and NeuroAI, investigating how experience shapes visual representations and recognition. My early work established the importance of viewpoint and learning in three-dimensional object recognition. With Isabel Gauthier, we used novel objects known as “Greebles”, demonstrating how perceptual expertise shapes responses in face-selective cortex, reframing debates about specialization in the human visual system.

My recent research uses machine learning to develop interpretable accounts of human vision. Our 2023 Nature Machine Intelligence study showed that visual representations learned through image-language training with CLIP better predict activity in particular subregions of higher visual cortex. Our BrainDiVE research combined models of brain responses with generative diffusion models to synthesize images predicted to strongly activate particular visual regions, with subsequent fMRI experiments testing these predictions directly. Complementary methods have described cortical preferences in natural language and identify the image content underlying neural responses. Alongside discoveries concerning food selectivity, our work reveals how visual features, semantic information, and behavioral relevance organize human visual cortex.

Throughout my career, I have integrated computational modeling with behavioral and neural evidence. My recent work uses naturalistic datasets and machine learning to connect findings within common computational frameworks and generate testable hypotheses. Contributions to fMRI analysis, models that generalize across individuals, and shared models of brain function extend this effort, linking questions in cognitive science with advances in artificial intelligence.

Other contributions include BOLD5000, the first public fMRI vision dataset; RSVP, an early free, open-source platform for behavioral research; and a longstanding repository of freely available visual stimuli. I also founded the Object Perception, Attention, and Memory (OPAM) conference.

how one person ended up working with me...

When people ask how I got into human-computer interaction, I tell them, "Well, originally I wanted to be a graphic designer, but a few serendipitous events changed my course. The first is that on my first day at college, in line at the computer store, I got into a conversation with a young guy with blue hair. It turned out he was a professor! He offered me a work-study job in his lab, and that opened my eyes to cognitive science. From there, one thing led to another…" 

— Scott Klemmer, Professor of Computer Science, UCSD (the blue-haired professor being me)