Multi-Feature Processing in Brains and Deep Neural Networks
In one of my central lines of work, I've been studying how visual systems, both biological and artificial, encode combinations of multiple visual features, focusing on color and shape representation. I have found that while the human visual system mostly tends to encode these variables in an orthogonal format, convolutional neural networks appear to process these features in an increasingly entangled manner over processing.
Taylor, J., & Karakose-Akbiyik, S. (2026). Reimagining the binding problem(s) for the 21st century. Symposium organized for the 26th annual meeting of the Vision Sciences Society Conference. St. Pete Beach, FL.
Xu, Y., Swinchoski, B., Taylor, J., & Chun, M. (2025). Interactive shape and color representation in visual working memory for colored objects in the human occipitotemporal cortex. Imaging Neuroscience, 3, IMAG-a.
Taylor, J., & Xu, Y. (2023). Comparing the Dominance of Color and Form Information across the Human Ventral Visual Pathway and Convolutional Neural Networks. Journal of Cognitive Neuroscience, 35(5), 816-840.
Taylor, J., & Xu, Y. (2022). Representation of color, form, and their conjunction across the human ventral visual pathway. Neuroimage, 251, 118941.
Taylor, J., & Xu, Y. (2021). Joint representation of color and form in convolutional neural networks: A stimulus-rich network perspective. PLOS One, 16(6), e0253442.
Taylor, J., & Xu, Y. (2022). Identifying the neural loci mediating conscious object orientation perception using fMRI MVPA. Cognitive Neuropsychology, 1-4.
Vaziri-Pashkam, M., Taylor, J., & Xu, Y. (2019). Spatial frequency tolerant visual object representations in the human ventral and dorsal visual processing pathways. Journal of cognitive neuroscience, 31(1), 49-63.
Taylor, J., & Xu, Y. (2019). The Coding of Color, Shape, and their Conjunction Across the Human Ventral Visual System. Talk presented at the 19th annual meeting of the Vision Sciences Society Conference. St. Pete Beach, FL.
NeuroAI Research Methods
I maintain an open-source Python package, TorchLens, for extracting, visualizing, and intervening on the internals of deep neural network models. It remains in active development and I welcome your thoughts on how to improve it.
Additionally, I've developed a method called framed RSA that incorporates mean activation information (e.g., the fact that a region responds more to faces than houses) into standard RSA, enabling richer adjudication of brain-computational models.
Taylor, J., & Kriegeskorte, N. (2025). Framed RSA: Representational comparisons that honor both geometry and population-mean response preferences. bioRxiv, 2025-07.
Taylor, J., & Kriegeskorte, N. (2023). Extracting and visualizing hidden activations and computational graphs of PyTorch models with TorchLens. Scientific Reports, 13(1), 14375.
Task-Relevant Visual Processing and Nonlinear Mixed Selectivity
In one line of work, I've studied how task modulates visual representations in the dorsal and ventral visual pathways. Specifically, I've developed a method called pattern-difference decoding that allows for non-invasively measuring nonlinear mixed selectivity (i.e., interaction effects) in the human brain using fMRI, finding evidence that the human dorsal stream multiplexes stimulus and task information in a nonlinear manner.
Taylor, J., & Xu, Y. (2024). Using fMRI to examine nonlinear mixed selectivity tuning to task and category in the human brain. Imaging Neuroscience, 2, imag-2.
Taylor, J., Vaziri-Pashkam, M., & Xu, Y. (2017). Effect of task on object category representations across human ventral, dorsal, and frontal brain regions. Poster presented at the 17th annual meeting of the Vision Sciences Society. St. Pete Beach, FL.
Taylor, J., Vaziri-Pashkam, M., & Xu, Y. (2016). Attention to Object Form Modulates Informational Connectivity Between Dorsal and Ventral Visual Streams. Poster presented at the 46th Annual Meeting of the Society for Neuroscience. San Diego, CA.
Face Perception
In past work with Gregory McCarthy, I studied how the electrophysiological markers of face perception varied between faces that were equated for their perceptual familiarity, but that differed in their level of associated semantic knowledge, finding that hallmarks of knowledge for face identity emerged as early as 200ms.
Taylor, J., Shehzad, Z., & McCarthy, G. (2016). Electrophysiological correlates of face-evoked person knowledge. Biological psychology, 118, 136-146.
Shehzad, Z., Taylor, J., & McCarthy, G. (2015). VATL Contributes to Biographical Knowledge of Faces via Feedback to FFA. Poster presented at the 45th Annual Meeting of the Society for Neuroscience. Chicago, IL.
