Hello there
Hello there
![copyright © YU Yue Me](/assets/me.c7c812f8.jpg)
About me
I am YU Yue (Will), an undergraduate at the University of Hong Kong.
Research
My research interest includes machine learning system, neural rendering, in-memory computing, and computer architecture.
NeRF
Neural radiance field (NeRF) implicitly represents a 3D scene with a multi-layer perceptron (MLP) for some position , view direction , color , and "opacity" . Rendered results are spectacular.
I am currently working on rapid and accurate 3D reconstruction of real-world scenes with NeRF. See my posts (link) on NeRF.
MLSys
![source: 10-414/714 by CMU needle](/assets/ndl.9339fc6f.png)
Needle is a deep learning library with customized GPU and NumPy CPU backend. The project is still in progress; see my repository for details.
[1]
To infinity and beyond![source: Dataset Distillation with Infinitely Wide Convolutional Networks by Timothy Nguyen, Roman Novak, Lechao Xiao, and Jaehoon Lee Dataset distillation](/assets/ds_distillation.3dbd4b2e.png)
Ever though of achieving % test accuracy on CIFAR-10 with merely data points? It is achieved with neural tangent kernel (NTK). This is the way.[2]
"To infinity and beyond" is a catchphrase by Buzz Lightyear from the Pixar animation film Toy Story. It is cited here to indicate NTK enables an infinitely wide hidden layer in neural networks. ↩︎
"This is the way" is a line from the Star Wars series the Mandalorian, which describes the "Mando" style. It is cited here to emphasize only NTK may achieve such an unbelievable result as dataset distillation. ↩︎