Zirui Chen

Krieger Hall 107

3400 N Charles St

Baltimore, MD 21218

I am a PhD student working with Dr. Michael F. Bonner, and I expect to finish in 2028. I am broadly interested in how visual systems, biological and artificial, come to represent the world the way they do. So far in my PhD, I have studied which parts of a visual representation are universal across these systems and which are idiosyncratic, how visual experience and learning objectives shape that split, and how much the answer depends on the measure used to compare them.

Prior to JHU, I received my bachelor’s degree from Emory University. I worked as a research assistant with Dr. Daniel Dilks on low-level visual features relevant to human visual scene processing.

Haven’t figured out what to share about my life outside of research. But cheers :D

news

May 18, 2026	I presented a poster on Learning dynamics reveal the natural basis of visual representation at VSS 2026.
Apr 30, 2026	We have a new preprint on Multidimensional dynamics of object representations in the human visual system.
Aug 13, 2025	I presented a poster on Neural dimensionality and temporal dynamics of visual representations at CCN 2025.
Jul 4, 2025	Our paper on Universal dimensions of visual representation is published in Science Advances.
May 19, 2025	I presented a poster on Unraveling the geometry of neural representational dynamics in rapid visual processing at VSS 2025.

selected publications

2025

Sci Adv

Universal dimensions of visual representation

Zirui Chen, and Michael F Bonner

Science Advances, 2025

Abs HTML PDF

Do neural network models of vision learn brain-aligned representations because they share architectural constraints and task objectives with biological vision or because they learn universal features of natural image processing? We characterized the universality of hundreds of thousands of representational dimensions from visual neural networks with varied construction. We found that networks with varied architectures and task objectives learn to represent natural images using a shared set of latent dimensions, despite appearing highly distinct at a surface level. Next, by comparing these networks with human brain representations measured with fMRI, we found that the most brain-aligned representations in neural networks are those that are universal and independent of a network’s specific characteristics. Remarkably, each network can be reduced to fewer than ten of its most universal dimensions with little impact on its representational similarity to the human brain. These results suggest that the underlying similarities between artificial and biological vision are primarily governed by a core set of universal image representations that are convergently learned by diverse systems.