Yixuan's bio photo

Yixuan (Sharon) Li

Pronounced as e-shwen

  G. Scholar LinkedIn Github e-Mail [CV]

About


I am currently a research scientist at Facebook AI. Before joining Facebook, I completed my PhD from Cornell University in December 2017, where I was advised by John E. Hopcroft. My thesis committee members are Kilian Q. Weinberger and Thorsten Joachims. I was honored to be selected as one of the Rising Stars in EECS 2017 by Stanford University in 2017.


My principal research interests are in machine learning and representation learning, with applications to computer vision and network science. I am excited about developing computational methods that can extract meaningful representations from masses of complex data, in forms of high-dimensional sensory inputs such as natural images, as well as structured data such as graphs.


In particular, my recent research focuses on designing machine learning models to enable more transparent, generalizable and robust representations for complex visual data, in settings where strong human supervision is either present or absent. Research topics that I'm working on, or have worked on, include:

  • Learning to generalize through collective intelligence: How can we train multiple learning agents and combine them to improve prediction?
  • Learning to generalize without strong supervision: How can we build models to learn representations that generalize to the open world, without using human annotation?
  • Learning to generalize without compromising robustness: How to make neural networks more robust to abnormal examples, such as out-of-distribution examples and adversarial examples?
  • Transparency of deep representations: How does a trained DNN encapsulate the structure that exists in a data set? How does learned representation differ across multiple independently trained networks? How does this guide improved design and training of more powerful networks?
  • Learning graph-structured representations and dynamics: How to design scalable algorithm to discover meaningful sub-structure in graphs with billions of nodes? How to model complex and possibly dynamic processes over large networks?

  • To tackle some of these research challanges, I've developed efficient and scalable machine learning methods that span topics on ensemble modeling, subspace learning, fast spectral clustering algorithms, semi-supervised learning and weakly supervised learning.



    I travel and occasionally take photos. Here is my pictorial Travel Memo.

    News


    12/13/2018: I will be giving a talk at Deep Learning Summit San Francisco in Jan 2019.
    10/1/2018: Gave an invited talk at Microsoft Research AI in Redmond, WA.
    9/27/2018: Served on a panel of Women in Research at Facebook.
    9/26/2018: Gave a talk at Grace Hopper Celebration (GHC) Artificial Intelligence track in Houston, TX.
    7/3/2018: Paper on exploring the limits of weakly supervised pretraining accepted into ECCV 2018.
    5/12/2018: Paper on understanding the loss surface of neural networks accepted into ICML 2018.
    4/3/2018: Received CVPR'18 Doctoral Consortium travel award.
    3/13/2018: Served on a panel at Facebook's Women in Research Lean In (WiRL) Circle.
    1/29/2018: Paper on Neural Network Reliability accepted into ICLR 2018.
    11/25/2017: Received ACM-W Scholarship in 2017.
    10/16/2017: I will be presenting at Women in Machine Learning (WiML) workshop in December this year.
    10/2017: Gave a talk at Grace Hopper Celebration (GHC) Artificial Intelligence track in Orlando, FL.
    8/5/2017: Selected as one of the Rising Stars in EECS 2017 by Stanford University.
    6/6/2017: Paper accepted for publication in Transactions on Knowledge Discovery from Data (TKDD).
    3/12/2017: Received ICLR 2017 Student Travel Award.
    2/27/2017: Paper on StackedGAN has been accepted into CVPR 2017.
    2/6/2017: Paper on Snapshot Ensembles has been accepted into ICLR 2017.