- 學位論文
漸變式資料概念飄移於實驗情境，資料，及模型學習之比較與探討
Incremental Data Drifting: Experiment Scenarios, Data, and Learning Model Comparisons
摘要
漸變式資料概念飄移是在將機器學習模型運用到真實世界時常見的問題。潛在的資料分布會與時遷移，例如：使用者於電商購物平台之購買喜好，會隨時間推移而改變。這個問題必須被深入探討，以達到良好的模型預測效果。現今研究漸變式資料概念飄移的文獻正面臨數個問題。第一，現階段缺乏清楚定義的漸變式概念飄移資料集。相關研究者僅限使用從過去真實資料搜集而來的資料集，或是人工生成資料集，兩者皆顯現其限制性。其中，前者缺乏明確定義的資料概念飄移種類及發生時間點。而後者，簡單的人工生成資料集，無法反應真實世界會遇到的資料複雜性。其二，目前尚未有一個完整衡量訓練模型於漸變式概念飄移資料集表現之協議。為了全面性探討漸變式資料概念飄移之議題，本作設計數方法以產生清楚定義之各類型漸變式概念飄移資料集。並且，本作設計ㄧ符合需求之衡量協議。此外，我們也探索機器學習其他領域，如域適應和終身學習，透過比較和實驗觀察這些方法在漸變式資料概念飄移之表現。
並列摘要
Incremental Data Drift is a common problem when employing a machine learning model to industrial applications. The underlying data distribution evolves gradually, e.g., users change their buying preference on an E-commerce website throughout time. The problem needs to be addressed to obtain high performance. Right now studies regarding Incremental Data Drift suffer from several issues. For one thing, there is a lack of clear-defined Incremental Drift datasets for examination. Researchers use either collected real-world datasets or synthesis datasets which show limitations. For the former one, in particular when and of which type of drifts the distribution undergoes is unknown, and for the latter one, a simple synthesis dataset cannot reflect the complex representation we would normally face in the real world. For another, there lacks a well-defined protocol to evaluate a learner's transfer abilities on an Incremental Drift dataset. To provide a holistic discussion on the issue, we create approaches to generate datasets of different drift types and define a protocol for evaluation. Also, we explore methods in other machine learning fields, such as Domain Adaptation and Lifelong Learning, to compare its performance on Incremental Data Drift.