Can Yaras


PhD Student
Electrical and Computer Engineering
University of Michigan - Ann Arbor
Email: cjyaras [at] umich [dot] edu

[Google Scholar][GitHub]

About Me

I am a PhD student in Electrical and Computer Engineering at the University of Michigan, advised by Qing Qu and Laura Balzano. I received my undergraduate degree from Duke University in Electrical and Computer Engineering, along with a major in Mathematics and a minor in Computer Science.

I am also a student researcher at Google NYC, hosted by Chong You.


My main research interests involve characterizing simple structures and dynamics in deep learning mainly through the lens of optimization, and applying those insights towards practical algorithms for efficient and scalable deep learning. I am also generally interested in nonconvex/manifold optimization.

Publications & Preprints

  1. C. Yaras*, P. Wang*, W. Hu, Z. Zhu, L. Balzano, Q. Qu. Invariant Low-Dimensional Subspaces in Gradient Descent for Learning Deep Matrix Factorizations. NeurIPS '23 M3L Workshop. [openreview]

  2. P. Wang*, X. Li*, C. Yaras, Z. Zhu, L. Balzano, W. Hu, Q. Qu. Understanding Deep Representation Learning via Layerwise Feature Compression and Discrimination. Preprint. [arxiv]

  3. C. Yaras*, P. Wang*, W. Hu, Z. Zhu, L. Balzano, Q. Qu. The Law of Parsimony in Gradient Descent for Learning Deep Linear Networks. Preprint. [arxiv] [github]

  4. C. Yaras*, P. Wang*, Z. Zhu, L. Balzano, Q. Qu. Neural Collapse with Normalized Features: A Geometric Analysis over the Riemannian Manifold. NeurIPS ’22. [arxiv] [github]

  5. P. Wang*, H. Liu*, C. Yaras*, L. Balzano, Q. Qu. Linear Convergence Analysis of Neural Collapse with Unconstrained Features. NeurIPS ’22 OPT Workshop. [opt-ml]

  6. T. Sarwar, C. Yaras, X. Li, Q. Qu, P.C. Ku. Miniaturizing a Chip-Scale Spectrometer Using Local Strain Engineering and Total-Variation Regularized Reconstruction. ACS Nano Letters. [pubs.acs]

  7. C. Yaras, K. Kassaw, B. Huang, K. Bradbury, J.M. Malof. Randomized Histogram Matching: A Simple Augmentation for Unsupervised Domain Adaptation in Overhead Imagery. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. [arxiv]