MIRlab - owen.lin RD page
本頁目錄
- Ref. http://mirlab.org/jang/mir/howToBuildPersonalRdPage.asp
- 說明：身為 MIRlab 多媒體資訊檢索實驗室的一員，會建立 RD page 置放個人研發相關資訊，如：讀過的論文、專案和連絡方式等。為避免放上實驗室主機後頁面難以維護，故透過 mirlab 網址導向至此，以利於維護之。
林育辰 Yu-Chen Lin
Education
- M. S., Department of Computer Science and Information Engineering, National Taiwan University, Sep. 2022~present
- B. S., Department of Computer Science and Information Engineering, National Taiwan Normal University, Sep. 2018~Jun. 2022
實驗室內職責
- NAS 機器管理
- 技術賦能組
- 語音組 --> NLP 組
Research
我目前在 Ansys 實習，主要研究大型語言模型的應用，如：LLM 於「專業領域」的 Q&A 問答，亦或是 工程領域上的程式碼補全等等，會預期會以 "Code Generator" 的方向為碩論主題。當前固定週三早上 Machine Learning Team Meeting 以及 每週五早上的 Weekly Meeting (我一人報告，與其他三四位專家討論)。若有新進展則會在 MIRlab 的 NLP Group 討論/分享。
以下我所閱讀的論文大多數以 自然語言處理 為主軸，未必涵蓋所有看過的論文，以下是我較有印象的。
Read papers
- Nye, Maxwell, et al. "Show your work: Scratchpads for intermediate computation with language models." arXiv preprint arXiv:2112.00114 (2021).
- Mohamed, Abdelrahman, et al. "Self-supervised speech representation learning: A review." IEEE Journal of Selected Topics in Signal Processing (2022). (Reading)
- Yao, Shunyu, et al. "React: Synergizing reasoning and acting in language models." arXiv preprint arXiv:2210.03629 (2022).
- Dua, Dheeru, et al. "Successive prompting for decomposing complex questions." arXiv preprint arXiv:2212.04092 (2022).
- Schick, Timo, et al. "Toolformer: Language models can teach themselves to use tools." arXiv preprint arXiv:2302.04761 (2023).
- Schäfer, Max, et al. "Adaptive test generation using a large language model." arXiv preprint arXiv:2302.06527 (2023).
- Zhao, Wayne Xin, et al. "A survey of large language models." arXiv preprint arXiv:2303.18223 (2023). (Reading)
- Peng, Baolin, et al. "Instruction tuning with gpt-4." arXiv preprint arXiv:2304.03277 (2023). (Reading)
- Park, Joon Sung, et al. "Generative agents: Interactive simulacra of human behavior." arXiv preprint arXiv:2304.03442 (2023).
- Chiang, Cheng-Han, and Hung-yi Lee. "Can Large Language Models Be an Alternative to Human Evaluations?." arXiv preprint arXiv:2305.01937 (2023).
- Chen, Lingjiao, Matei Zaharia, and James Zou. "FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performance." arXiv preprint arXiv:2305.05176 (2023).
- Wang, Yue, et al. "Codet5+: Open code large language models for code understanding and generation." arXiv preprint arXiv:2305.07922 (2023). (Reading)
- Wang, Peiyi, et al. "Large language models are not fair evaluators." arXiv preprint arXiv:2305.17926 (2023).
- Thakur, Shailja, et al. "VeriGen: A Large Language Model for Verilog Code Generation." arXiv preprint arXiv:2308.00708 (2023).
- Zheng, Lianmin, et al. "Judging LLM-as-a-judge with MT-Bench and Chatbot Arena." arXiv preprint arXiv:2306.05685 (2023).
- He, Zhuolun, et al. "ChatEDA: A Large Language Model Powered Autonomous Agent for EDA." arXiv preprint arXiv:2308.10204 (2023).
- Sun, Mingjie, et al. "A Simple and Effective Pruning Approach for Large Language Models." arXiv preprint arXiv:2306.11695 (2023). (Reading)
Read articles
這裡指的是普通的技術文章，而不一定是指學術專精的 article
- 淺談神經機器翻譯 & 用 Transformer 與 TensorFlow 2 英翻中
- Self-supervised learning and its applications to speech processing(李宏毅教授短講)
Read video- 機器學習 2023 - 生成式AI 系列影片
Read video - 宏毅老師系列
- 【機器學習2021】生成式對抗網路 (Generative Adversarial Network, GAN) (一) – 基本概念介紹
- 【機器學習2021】生成式對抗網路 (Generative Adversarial Network, GAN) (二) – 理論介紹與WGAN
- 【機器學習2021】自督導式學習 (Self-supervised Learning) (一) – 芝麻街與進擊的巨人
- 【機器學習2021】自督導式學習 (Self-supervised Learning) (二) – BERT簡介
- 【機器學習2021】自編碼器 (Auto-encoder) (下) – 領結變聲器與更多應用
- 【機器學習2021】來自人類的惡意攻擊 (Adversarial Attack) (下) – 類神經網路能否躲過人類深不見底的惡意？
- 【機器學習2021】機器學習模型的可解釋性 (Explainable ML) (上) – 為什麼類神經網路可以正確分辨寶可夢和數碼寶貝呢？
- 【機器學習2021】機器學習模型的可解釋性 (Explainable ML) (下) –機器心中的貓長什麼樣子？
- 【機器學習2021】概述領域自適應 (Domain Adaptation)
- 【機器學習2021】概述增強式學習 (Reinforcement Learning, RL) (一) – 增強式學習跟機器學習一樣都是三個步驟
- Transformer
- ELMO. BERT, GPT (Reading)
Competition
- 最佳成績：第一名, 臺大「深度學習之應用」期末 Hahow 課程預測競賽
- 優勝 (第六名), 玉山人工智慧公開挑戰賽 2022 夏季賽 —— 語音辨識後修正, 林育辰, 梁俊彥
Project
- 字幕生成系統 (A Subtitles Generator using WeNet Toolkits)
- 輸入 YouTube 網址可抽取音訊，生成其對應字幕之展示系統
- 於 DSP 課程中獲 A+ 成績，其持續衍生開發，目前為 MIRlab 展示成品之一
Meeting Slide
- [Password] [Slide Folder]
- Note: Password 限內部有帳號者才能存取，得到 Password 後，可進入 slide folder 鍵入 password 即可觀看全年的報告投影片。
其他
- 手冊：[技術賦能組], [語音組手冊] (手冊大部分由我建置與撰寫，若需連結請私訊或翻 LINE 群組記事本)
Course
碩士
- 機器學習 (Machine Learning, ML), 李宏毅老師
- 深度學習之應用 (Applied Deep Learning, ADL), 陳縕儂老師
- 數位語音處理概論 (Introduction to Digital Speech Processing, DSP), 李琳山老師
- 自然語言處理 (Natural Language Processing, NLP), 陳信希老師
- 高等電腦視覺 (Advanced Computer Vision, ACV), 傅楸善老師
- 智慧型汽車導論 (Introduction to Intelligent Vehicles), 林忠緯老師
- 類神經網路 (Neural Network, NN), 劉長遠老師
- 腦理論 (Brain Theory), 劉長遠老師
大學
- 資料探勘、資料視覺化、計算機圖學、影像處理
- 人工智慧、啟發式演算法與解題應用
- 區域性網路、資料通訊
- 資料庫理論
- 生物技術、生物資訊應用程式語言、生物資訊學導論
- 作業系統、計算機結構、電腦輔助VLSI設計、組合語言、數位邏輯
- 程式設計（一）（二）、進階程式設計
- 演算法、資料結構
- 機率論、線性代數、離散數學、微積分乙（一）（二）
- ..., 共計：150.0 學分
Contact
- Ref. http://mirlab.org/jang/mir/howToBuildPersonalRdPage.asp
- About this page: As a member of MIRlab, the Multimedia Information Retrieval Laboratory, I maintain an R&D page with information about my research and development work, including papers I have read, projects, and contact details. To make the page easier to maintain than one hosted on the laboratory's server, my MIRlab URL redirects here.
Yu-Chen Lin 林育辰
Education
- M. S., Department of Computer Science and Information Engineering, National Taiwan University, Sep. 2022~present
- B. S., Department of Computer Science and Information Engineering, National Taiwan Normal University, Sep. 2018~Jun. 2022
Laboratory Responsibilities
- NAS server administration
- Technical Enablement Team
- Speech Group --> NLP Group
Research
I am currently interning at Ansys, where my research focuses on applications of large language models, such as question answering in specialized domains and code completion for engineering applications. I expect to focus my master's thesis on a "Code Generator". I attend a Machine Learning Team Meeting every Wednesday morning and a Weekly Meeting every Friday morning. At the Friday meeting, I am the sole presenter and discuss my work with three or four experts. When I make new progress, I share and discuss it with MIRlab's NLP Group.
Most of the papers I read focus on natural language processing. The list below is not an exhaustive record of everything I have read; it includes the papers that stand out to me.
Papers
- Nye, Maxwell, et al. "Show your work: Scratchpads for intermediate computation with language models." arXiv preprint arXiv:2112.00114 (2021).
- Mohamed, Abdelrahman, et al. "Self-supervised speech representation learning: A review." IEEE Journal of Selected Topics in Signal Processing (2022). (Reading)
- Yao, Shunyu, et al. "React: Synergizing reasoning and acting in language models." arXiv preprint arXiv:2210.03629 (2022).
- Dua, Dheeru, et al. "Successive prompting for decomposing complex questions." arXiv preprint arXiv:2212.04092 (2022).
- Schick, Timo, et al. "Toolformer: Language models can teach themselves to use tools." arXiv preprint arXiv:2302.04761 (2023).
- Schäfer, Max, et al. "Adaptive test generation using a large language model." arXiv preprint arXiv:2302.06527 (2023).
- Zhao, Wayne Xin, et al. "A survey of large language models." arXiv preprint arXiv:2303.18223 (2023). (Reading)
- Peng, Baolin, et al. "Instruction tuning with gpt-4." arXiv preprint arXiv:2304.03277 (2023). (Reading)
- Park, Joon Sung, et al. "Generative agents: Interactive simulacra of human behavior." arXiv preprint arXiv:2304.03442 (2023).
- Chiang, Cheng-Han, and Hung-yi Lee. "Can Large Language Models Be an Alternative to Human Evaluations?." arXiv preprint arXiv:2305.01937 (2023).
- Chen, Lingjiao, Matei Zaharia, and James Zou. "FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performance." arXiv preprint arXiv:2305.05176 (2023).
- Wang, Yue, et al. "Codet5+: Open code large language models for code understanding and generation." arXiv preprint arXiv:2305.07922 (2023). (Reading)
- Wang, Peiyi, et al. "Large language models are not fair evaluators." arXiv preprint arXiv:2305.17926 (2023).
- Thakur, Shailja, et al. "VeriGen: A Large Language Model for Verilog Code Generation." arXiv preprint arXiv:2308.00708 (2023).
- Zheng, Lianmin, et al. "Judging LLM-as-a-judge with MT-Bench and Chatbot Arena." arXiv preprint arXiv:2306.05685 (2023).
- He, Zhuolun, et al. "ChatEDA: A Large Language Model Powered Autonomous Agent for EDA." arXiv preprint arXiv:2308.10204 (2023).
- Sun, Mingjie, et al. "A Simple and Effective Pruning Approach for Large Language Models." arXiv preprint arXiv:2306.11695 (2023). (Reading)
Technical Articles
These are general technical articles, rather than necessarily scholarly articles.
- An Introduction to Neural Machine Translation: Translating English into Chinese with Transformer and TensorFlow 2
- Self-supervised Learning and Its Applications to Speech Processing (A Short Lecture by Prof. 李宏毅)
Videos: Machine Learning 2023 — Generative AI Series
Videos: Prof. 李宏毅's Series
- Machine Learning 2021: Generative Adversarial Networks (GAN), Part I — Basic Concepts
- Machine Learning 2021: Generative Adversarial Networks (GAN), Part II — Theory and WGAN
- Machine Learning 2021: Self-supervised Learning, Part I — Sesame Street and Attack on Titan
- Machine Learning 2021: Self-supervised Learning, Part II — An Introduction to BERT
- Machine Learning 2021: Autoencoders, Part II — The Bow-tie Voice Converter and More Applications
- Machine Learning 2021: Adversarial Attacks, Part II — Can Neural Networks Escape Humanity's Boundless Malice?
- Machine Learning 2021: Explainable ML, Part I — Why Can Neural Networks Correctly Distinguish Pokémon from Digimon?
- Machine Learning 2021: Explainable ML, Part II — What Does a Cat Look Like to a Machine?
- Machine Learning 2021: An Overview of Domain Adaptation
- Machine Learning 2021: An Overview of Reinforcement Learning, Part I — The Same Three Steps as Machine Learning
- Transformer
- ELMO. BERT, GPT (Reading)
Competitions
- Best result: First place in the Hahow course prediction competition, the final project for NTU's Applied Deep Learning course.
- Prize winner (sixth place) in the E.SUN AI Open Challenge 2022 Summer Competition — Speech Recognition Post-correction, with team members Yu-Chen Lin (林育辰) and 梁俊彥.
- Competition notes: https://hackmd.io/@mathlin/BJ4J1Nuc5
- Reflection article: https://reurl.cc/1mKoOY
- GitHub: https://github.com/yuchen0515/2022-Competition-CUDAOutOfMemory
Project
- Subtitle Generation System (A Subtitles Generator using WeNet Toolkits)
- A demonstration system that takes a YouTube URL, extracts the audio, and generates corresponding subtitles.
- The project received an A+ in the DSP course. Development has continued, and it is currently one of MIRlab's demonstration projects.
Meeting Slides
- [Password] [Slide Folder]
- Note: The password is accessible only to internal users with an account. After retrieving it, open the slide folder and enter the password to view the entire year's presentation slides.
Other Resources
- Handbooks: [Technical Enablement Team], [Speech Group Handbook]. I created and wrote most of these handbooks. For the links, please message me privately or check the notes in the LINE group.
Coursework
Master's Courses
- Machine Learning (ML), Prof. 李宏毅
- Applied Deep Learning (ADL), Prof. 陳縕儂
- Introduction to Digital Speech Processing (DSP), Prof. 李琳山
- Natural Language Processing (NLP), Prof. 陳信希
- Advanced Computer Vision (ACV), Prof. 傅楸善
- Introduction to Intelligent Vehicles, Prof. 林忠緯
- Neural Network (NN), Prof. 劉長遠
- Brain Theory, Prof. 劉長遠
Bachelor's Courses
- Data Mining, Data Visualization, Computer Graphics, Image Processing
- Artificial Intelligence, Heuristic Algorithms and Problem-solving Applications
- Local Area Networks, Data Communications
- Database Theory
- Biotechnology, Programming Languages for Bioinformatics, Introduction to Bioinformatics
- Operating Systems, Computer Architecture, Computer-aided VLSI Design, Assembly Language, Digital Logic
- Programming I and II, Advanced Programming
- Algorithms, Data Structures
- Probability Theory, Linear Algebra, Discrete Mathematics, Calculus B I and II
- ..., 150.0 credits in total
Contact
一起討論
歡迎分享想法、補充資料，或留下你的問題。使用 GitHub 帳號登入後，即可在這裡留言。
留言將在這裡載入。
查看留言原始討論