About Me

Personal Information¶

• Research Interests
  • Machine Learning
  • Deep Learning
  • Computer Vision
  • Audio Recognition
  • Natural Language Processing
  • Social Network Analysis

Competitions¶

• 電機系計算機概論盃 Blokus 策略程式設計競賽 季軍 (2020-05)

• 大專校院資訊應用服務創新競賽 佳作 (2020-10)
  We developed a route-recommendation system considering factors including but not limited to traffic, weather, disease(登革熱), user preference.
  In this project, I was in charge of designing and implementing algorithm for shortest path problem, and data collecting and visualizing data for 登革熱.

• Kaggle – BirdClef (2022-03)
  BirdClef, audio classification task, aims to distinguish between bird chirping of various species. I was not familiar to neither NLP nor audio recognition before participating this competition. At that time, I simply adapted ViT (Vision Transformer) to audio version and thus only got average grade. However, I started to study in NLP and audio recognition and further influenced my personal research Music Popularity Prediction.
• AI CUP 2022 肺腺癌病理切片影像之腫瘤氣道擴散偵測競賽 II：運用影像分割作法於切割STAS輪廓 (2022-04) ，獲得前標 14%
  Stas is an image semantic segmentation task with only two classes which are normal cells and abnormal cells. The difficulty lies in the fact that even untrained human is hard to distinguish the labels correctly.
  In this competition, I adapted SegFormer and proposed the method Spot Validation validating ROI with more robust CNN model instead of using the output of SegFormer directly.

Projects¶

Music Popularity Prediction (Jul, 2022 ~ Present)¶

This project/research aims to predict the time-series popularity of music with transformer architecture, and is conducted, researched, developed by myself.

See hackmd for more details.

• 中文：https://hackmd.io/4pdMiliQSHaS5jb6J5BaDg
• English: https://hackmd.io/W03TtVhOSQCbceFOnd5vNQ

Ahk-Doc (Feb, 2022 ~ Present)¶

Ahk-Doc is a language support service for AutoHotkey which is an unpopular script language and thus the original language support lacked many useful features.

Currently, I have been working on refactoring for support of LSP(Language Server Protocol) which can not only make language server available to other popular IDE but also improve performance.

The original project (which I forked from) only used several regular expression as parser whereas it cannot work on complicated expression and thus make some advanced services unavailable. After studying the theorem of compiler, I'm going to implement more robust lexer and parser for the important but still lacking features of language service.

Minor¶

• pytorch-model-utils – toolkits based on Pytorch
• Lernen – Social Website for learner.

Skills¶

• Programing Languages
  • Python Proficient
  • Javascript / Typescript Proficient
  • C++ Competent
  • C Competent
  • C# Advanced Beginner
  • Java Advanced Beginner
  • Autohotkey Expert
• Web
  • Vue2 Competent
  • Vue3 Advanced Beginner
  • Express (node.js) Competent
  • Database Management Competent

Grades¶

Honors & Awards¶

• Contributed to Pytorch

Study Plan¶

Realized the most important parts are those basic or traditional theories, in addition to continue my working projects and research, I am studying data compression (highly related to information theory and signal processing), statistics, digital speech processing, and so on.

Hobby¶

• Badminton
  Playing badminton at least once a week for cultivation of spirit and winding down from pressure.