Yufan He

Contact: heyufan1995@gmail.com

I am currently an applied research scientist at NVidia, and I'm working on AutoML, transformers, self-supervised learning, and Project MONAI.

Before that, I received my M.S. and Ph.D. from the Johns Hopkins University and my B.S. from Tsinghua University.

[View my Resume]

Education

Johns Hopkins University

Master of Science
Doctor of Philosophy
Electrical and Computer Engineering
2016 - 2021

Tsinghua University

Bachelor Degree
Electronic Engineering
Economics and Management (Second Degree)
2012 - 2016

Publications

This section contains part of projects that I have participated. Visit my Google Scholar for a full list of papers

DiNTS: Differentiable Neural Network Topology Search for 3D Medical Image Segmentation. (CVPR 2021 oral)

A differentiable neural archtecture search method for 3D medical image segmentation. It contains a new search space and a novel multi-path relaxation scheme for differentiable search. [code] [paper].

TransMorph: Transformer for unsupervised medical image registration. (submitted to Medical Image Analysis)

A state-of-the-art unsupervised 3D medical image registration package using swin transformer. [code] [paper].

An ealier conference version "ViT-V-Net: Vision Transformer for Unsupervised Volumetric Medical Image Registration", MIDL 2021, is available [code] [paper].

Self domain adapted network. (MICCAI 2020 oral)

A test-time self domain adaptation method is proposed. It uses auto-encoders' reconstruction loss as a source-target alignment measurement. A set of adaptors are trained and used for domain adaptation in test-time. [code] [paper].

The journal version "Autoencoder based self-supervised test-time adaptation for medical image analysis", Medical Image Analysis 2021, is available [paper].

Validating uncertainty in medical image analysis. (ISBI 2020 & SPIE2020)

Two papers about estimating and utilizing the uncertainty in medical image translation. [code] [ISBI paper] [SPIE paper].

Fully Convolutional Boundary Regression for Retina OCT Segmentation. (MICCAI 2019 oral)

A novel fully convolutional boundary regression method is proposed for efficient, end-to-end, and topology guaranteed surface segmentation. [code] [paper].

The journal version "Structured layer surface segmentation for retina OCT using fully convolutional regression networks", Medical Image Analysis 2020, is also available [paper].

Experience

Research Assistant

IACL, Johns Hopkins University, Sep.2016 - Nov.2021

PICSL, University of Pennsylvania, June 2015 - Sep 2015

Teaching Assistant

Compressed Sensing, Spring 2018 & 2019, Johns Hopkins University

Random Signal Analysis, Fall 2017, Johns Hopkins University

Internship

AutoML in Medical Imaging, June 2020 - Sep 2020

Autonomous Driving, July 2018 - Sep 2018

Awards

IEEE TMI Distinguished Reviewer

MICCAI 2020 Student Travel Award

MICCAI 2019 Young Scientist Award

MICCAI 2019 Student Travel Award

The 2018 George M.L. Sommerman Engineering Graduate Teaching Assistant Award

4th MICCAI Workshop on Ophthalmic Medical Image Analysis 2017, Best Oral Paper Award

Miscellaneous

My lovely dog Kunkka