Stella Yu



Publications

stellayu @ berkeley . edu

I am the director of the Vision Group at the International Computer Science Institute, a research institute affiliated with UC Berkeley.

I am a Senior Fellow at the Berkeley Institute for Data Science, a central hub of research and education at UC Berkeley designed to facilitate and nurture data-intensive science.

As a vision scientist, I subscribe to the view that visual perception presents not just a fascinating computational problem, but more importantly an intelligent solution for large-scale data mining and pattern recognition applications. I study visual perception from multiple perspectives: human vision, computer vision, robotic vision, and artistic vision.

I am particularly interested in human vision's ingenuity and immediacy at resolving a property or region of interest from its context. The context is not a mere conglomeration of pieces, but an organized whole of interrelated elements. Only in such an organization can an object be assigned a place and be singled out from the context without relinquishing the effects of the context.

My approach to computer vision is thus about establishing parts in relation, organizing them into a whole, and singling out the aspect of interest from the whole. In modeling visual organization, I have developed new tools in spectral graph theory and machine learning to effectively uncover the relational structure of large-scale data. I have also pursued efficient representation and computation by learning from artists and exploring the connections between Art and Vision.

Human Vision

brightness perception

scene categorization

change blindness

understanding popout

saccadic suppression

Computer Vision

object matching

image segmentation

finding dots and tubes

inferring spatial layout

object segmentation

Machine Learning

angular embedding

multiclass spectral clustering

grouping with bias

grouping with repulsion

Markov decision process