Why use kernel size of 4 in discriminators?
Published:
Why use kernel size of 4 in discriminators? Because of Checkboard artifacts in gradients Link
Posts by Tags
Checkboard artifacts
Generative Adversarial Net
Why use kernel size of 4 in discriminators?
Published:
Why use kernel size of 4 in discriminators? Because of Checkboard artifacts in gradients Link
GAN paper reading
Published:
Progressive GAN
Progressive growing of GAN
- Start with low-resolution images and then progressively increasing the resolution by adding layers to the network!
Greatest mind of all time
Quotes, by Yuan Cui and Shi Su
Published:
古之所谓豪杰之士，必有过人之节。人情有所不能忍者，匹夫见辱，拔剑而起，挺身而斗，此不足为勇也。天下有大勇者，卒然临之而不惊，无故加之而不怒。此其所挟持者甚大，而其志甚远也。
Keyword Spotting
Keyword Spotting with hardware constraint
Published:
This is a slide about Keyword Spotting with hardware constraint. Some model compression approaches are also introduced to reduce the computational requirement.
Learning
Model Compression
Keyword Spotting with hardware constraint
Published:
This is a slide about Keyword Spotting with hardware constraint. Some model compression approaches are also introduced to reduce the computational requirement.
Music
Omnipotent Youth Society
Published:
Omnipotent Youth Society (万能青年旅店，简称万青), Chinese indie band in Hebei. blues rock & indie rock.
Lo Tayou (罗大佑), Chinese Musician born in Taiwan.Teaching
Teaching Visualizations
Published:
To facilitate the teaching of a certain undergraduate class, I created some visualizations for several basic machine learning algorithms, including K-Means clustering, e.g. These visualizations help students to better understand the underlying concepts of these algorithms and how they work in practice.