嗨,我是楊凱州,目前是個平凡的研究生,正在探索人工智慧與資料科學的藝術,期許自己有朝一日能成為 Kaggle 與 Gym 的大魔王。 我喜歡關注各式各樣的開源會議,熱衷於嘗試一些天馬行空的發想,藉由開發一些有趣的小專案自娛娛人,同時也提高自己的實務水平。
我也是雷德麥的藏書閣與 NCKUCSIE Notes 的作者, 閒暇之餘會寫些技術報告或學習心得。 我相信技術是為了生活而存在,而致力於用說故事的方式, 搭配上生活化例子擊破那些艱澀的理論。 若您在文章的實作上遭遇到了困難,還是發現內容存在有待加強的地方, 都歡迎藉由 G-mail 與我分享您的觀點,期待透過討論能讓彼此更加強大 ヾ(´︶`*)ノ
Department of Computer Science• July 2013 - July 2017
伴我從新手村走到二轉之路的好夥伴。 一路上挑戰了形形色色的演算法與資料結構,從中認識了不少程式語言,建構出許許多多妙趣橫生的專案。 觀過網路架構的層巒疊嶂,體會過模糊理論的道不清說不明,玩轉過了一道又一道的密碼矩陣, 還有更多更多沒提到的精彩冒險,且讓我日後在藏書閣中娓娓道來吧。
2nd place• August 2018
This competition is about short-text semantic similarity and cross-lingual knowledge transfer. I design three deep learning models based on different hypotheses and create the variety intra-model by mixing various views of input data.
3rd place• February 2019
This is a challenge for Chinese fake-news detection. I implemented various NLI networks and inject the world knowledge using BERT and proposed a rumor-aware model which is inspired by decomposable attention.
4th place• April 2019
Pronoun resolution is part of coreference resolution, the task of pairing an expression to its referring entity. In this task, I conducted extensive analysis among hidden layers of BERT and proposed a multi-head siamese scorer for answer selection.
top 1%• May 2018
Aim at predicting concentration levels of PM2.5, PM 10 and O3 over the coming 48 hours for Beijing and London. I reach top 1% without any external data by an end to end sequence generative network that fuses the views from regional and sequential information efficiently.
top 1%• February 2018
A challenge for multi-label sentence classification. To identify the toxic comments, I present two simple and easy-to-use architectures for generating reliable sentence embeddings and ensemble them with gradient boosting trees.
top 1%• November 2017
The goal is to predict the chances of a user listening to a song repetitively after the first observable listening event within a time window was triggered. I model the user and song pair by deep factorization machine and gradient boosting trees with effective feature engineering.
兼任助理• July 2015 - August 2015
使用 Smarty 與 JQuery 維護成功大學醫學院 e 化學習系統,並針對現有的 Bug 進行除錯。 以及協助進行農業地理資訊統計,調用 GoogleMap API 進行地圖打點和編寫地址至經緯度的批量轉換器。
研究員• July 2016 - February 2017
執行科技部大專生研究計畫,實作具即時語音信息擷取功能之緊急任務救助服務匹配系統,採用自然語言處理技術,抽取短語中的潛在意圖,並透過定義缺漏特徵的個數,以構件多輪式的對話情境。
資訊服務人員• March 2015 - July 2015
以自身技術協助其他社會福利機構,如 Excel 的表單統計與函式除錯,個人在程式上的學習經驗分享等等。
採用爐石戰記佈景的西洋棋遊戲,使用遊戲樹配合 Alpha-Beta 剪枝來實現人機對戰,目前已經可以前瞻後 8 手的盤面。
Java, 人工智慧, 遊戲設計Top 1% solution to the toxic comment classification challenge on Kaggle
深度學習, 文本分類以詞嵌入為基底,開發樣板式、檢索式及生成式的聊天機器人,透過上層分類器的可靠度比對來決定回應。 此外也針對樣板規則定義主題,實作出對應的主題模組,用來處理多輪式對話。
Python, 自然語言處理, 模糊比對, 機器學習