> Machine Learning

Name: Air-Quility Prediction

Demo: https://aq.zio.tw/ (offline)

Description: The topic of KDD-CUP 2018 contest. Predict both Beijing and London air-quality value including CO2, NO, NO2 and so on. Base on given datasets from the contest, preprocessing and selecting features, using Pytorch to define LSTM model, predict the next 48 hours air quality daily. The grade is calculated base on the average score for the whole month.

Result: KDD CUP #113

Name: Wounds images classfication

Demo: https://wounds.zio.tw/ (offline)

Description: It’s a project that I cooperate with another department. Because dataset is so small that I can not train model base on CNN from scratch, I use fine-tune methods to reach an approximately better accuracy of prediction 70%. BUT, our objective is to reach 90% above, so I still find a way to upgrade our model and develop new methods to increase accuracy. As so far, I am studying in the one-shot problem of machine learning to find a suitably state-of-the-art method to solve this problem.

Name: Wounds images GAN

Description: It’s a project that I use for data augmentation for the problem just mentioned. I combine two paper methods to define a whole model. First, I made a AC-GAN for a basic GAN model to generate images according the class. Second, because small data problem, it is easy to make the Discriminator too strong that model collapse. To deal with that, I add the limitation call Spectral Normalization from this paper to ensure our quality of productions from Generator are good enough.


Name: 黑函

Link: https://iorz.cc/

Description: It is my personal blog for technique-oriented. Using Hugo static site generator.

Deploy service: Netlify

Name: アイネクライネ

Link: https://murmur.zio.tw/

Description: It is my personal blog for mood-oriented. Using Hugo static site generator.

Deploy service: Netlify

Name: Comment System

Link: https://murmur.zio.tw/comment

Description: It is a static site comment system for my blog. Wrapping it with a serverless architecture used for api. Fronted end side use vanilla Javascript to render. Using webpack to group all components.

Deploy service: Now.sh

Name: ThisIsOrz

Link: https://iorz.cc/thisisorz/

Description: A wordpress Theme.

Name: Petime

Description: A service developed by MEAN stack (MongoDB + Express + Angular 1 + Node.js), offer a platform for who has requirements that feeder needs someone to take care of his pets temporary.

All source code is serve on Github. There are some other projects not mentioning above. Here is my Github (But a lot of projects is running on private repository. XD)