Meenakshi Khosla- Understanding Vision Through Neural Supervision and Generative Probes

Understanding Vision Through Neural Supervision and Generative Probes

Abstract

How can neural data guide the learning of visual representations, and how can we choose stimuli that reveal their properties? I will present two lines of work addressing these questions. First, I will show how training a model to emulate V4 responses yields visual representations that capture multiple signatures of human perception, often better than task-optimized models, highlighting what biological supervision contributes beyond task-based training. Second, I will introduce diffusion models as stimulus engines for probing sensitivity in biological and artificial visual systems. We use the local geometry of a learned natural-image distribution to construct controlled image perturbations, revealing a systematic relationship between density curvature and human and model sensitivity. This approach makes the choice of where to probe stimulus space an explicit part of understanding and comparing visual representations. Together, these directions point toward a productive exchange between neuroscience and AI: neural data can guide more human-like models, while generative models open new avenues for discovering how biological and artificial visual systems represent the world.

Meenakshi Khosla
 

UC San Diego, Department of Cognitive Science

Meenakshi Khosla is an Assistant Professor of Cognitive Science at UC San Diego, with an affiliate appointment in Computer Science and Engineering. Her research at the intersection of computational neuroscience and machine learning aims to understand how brains and artificial neural networks represent information and how these representations support perception and cognition. She received her Ph.D. in Electrical and Computer Engineering from Cornell University and was an ICoN Postdoctoral Fellow at MIT. Her research is supported by the Kavli Institute for Brain and Mind, Coefficient Giving (formerly Open Philanthropy), and NSF’s Collaborative Research in Computational Neuroscience grants.

https://cogsci.ucsd.edu/people/faculty/meenakshi-khosla.html
 

About the Center for Neural Data Science Seminar Series 

The Center for Neural Data Science Seminar Series is a platform for trainees across campus to share  insights and innovative approaches that bridge the gap between neuroscience and data science. 

As neuroscience continues to generate vast amounts of data—from intricate neural circuit maps to large-scale brain activity recordings—the need for interdisciplinary expertise in data science, statistics, and engineering has never been more critical. 

The Center for Neural Data Science mission is to advance brain research through the development of cutting-edge analytical methodologies and collaborative approaches. Stanford's affiliates are invited to join this vibrant community dedicated to transformative discoveries.

This seminar series is only offered in person. 

Happy Hour/Reception for Attendees: 

Seminars attendees are invited to stay for a happy hour reception after the seminar to continue the conversation with the speaker.

Sign up to learn about all our upcoming events