My name is Toon Tran (yes as in car-TOON) or Tung Tran. I double major in Math and Computer Science at Bucknell University.

I’m fortunate to be advised by the wonderful Professor Joshua Stough (Bucknell University) and Dr. Haggerty (Geisinger Health) in my previous research on automatic echocardiography segmentation.

[CV] [Google Scholar] [Github]
Email: tst008 [at] bucknell [dot] edu

Research

My current interest lies in machine learning, optimization and deep learning.

Specifically, I’m looking at theoretical deep learning frameworks such as Tensor Programs, Neural Tangent Kernels, or data-dependent two-layer networks to approach memorization in neural networks. Memorization is referred to here as both tendency to memorize noise (overfit) and statistical features (causing lower performance on distribution shift).

Previously, I did research on language models for theorem proving. Specifically, I examined the ability of the neural network GPT-2 to incrementally learn to prove math statements from its own mistakes in Lean, an Interactive Theorem Prover for doing math on computers. Finding my model does not respond well to distribution shift, I decided to dive into the fundamentals of deep learning instead. This topic is still in my interest, though I will pursue it at a later time.

In another previous research, I looked into automatic echocardiography segmentation using CNNs. My model also exhibit memorization, making it hard to generalize across different datasets.

Publications

Conference Papers

Bayesian Optimization of 2D Echocardiography Segmentation
Toon Tran, Joshua V. Stough, Xiaoyan Zhang, Christopher M. Haggerty
In Proceedings of the IEEE International Symposium on Biomedical Imaging (ISBI ‘21)
[arxiv] [poster] [video]

Technical Reports and Preprints

Symmetric Diffeomorphic Registration as Segmentation Post-processing Method
Toon Tran, Joshua V. Stough
Technical Report
[pdf]

Miscellaneous

• I enjoy reading comics, fantasy and sci-fi books. I also play badminton casually when I get the chance.
• If you want to discuss or study epistemology/philosphy of science, do hit me up.
• Please feel free to schedule a meeting or send me an email if you have any question about anything, or if you just want to have a chat!