- Curiosity and Enthusiasm -
Welcome to Chengyu Dong’s homepage! My name in Chinese is 董(supervise)城(city wall)昱(sunlight ☀️). I am an enthusiastic researcher in machine learning, particularly interested in robust learning and few-shot learning. My vision is to uncover the learning process in a principled way. I was a researcher in astronomy, interested in solar system dynamics.
I love soccer ⚽. I used to play every week before the pandemic era. My favourite club is Dortmund. DM me if you want to watch soccer games together 🐶!
M.S. in Computer Science, 2020
University of California, San Diego
M.S. in Astrometry and Celestial Mechanics, 2018
B.S. in Astronomy, 2016
Multiple problems in adversarial training including robustness-accuracy trade-off, robust overfitting, and gradient masking share one commnon cause – low quality samples in the dataset.
The average of contextualized representations shares almost the same direction as the first principal component of the matrix whose columns are these representations. We believe this explains why the average representation is always a simple yet strong baseline.
We propose APART, an adaptive adversarial training framework, which parameterizes perturbation generation and progressively strengthens them.
We motivate from Neural-ODE perspective and design an adaptive training algorithm for ResNet, which can save ~50% training time.
We estimate the frequency of close encounters and collisions between Plutinos and Nepture Trojans.
The extension of the CH4 torus around Titan depends critically on the population of the high-energy tail of the CH4 energy distribution.