電機資訊學院: 資訊工程學研究所指導教授: 林軒田尤聖棨You, Sheng-ChiSheng-ChiYou2017-03-032018-07-052017-03-032018-07-052015http://ntur.lib.ntu.edu.tw//handle/246246/275541現實生活中的線上學習問題，通常伴隨著隨時改變的學習目標函式。這種改變「觀念偏移」，通常會降低線上學習演算法的效能。因此許多研究針對此嘗試分析資料的統計特性、提出偵測偏移改變的門檻機制。或者直接應用框格窗的方法維護較新的資料集合，反應出最近資料的觀念偏移。然而，鮮少研究同時專注於「偵測」偏移並維護「部份」資料來改善學習演算法。我們提出了一個基於現有的線上學習演算法的框架，在觀念偏移的資料提昇效能。此框架利用檢查抑制更新資料點（反更新）來檢查是否為觀念偏移，讓原算法得到更好的學習效果。此框架會根據套用的學習算法和處理的資料特性，讓學習演算法自然地產生動態框格窗來維持部份資料以對應資料觀念偏移的走勢。我們提出數種不同的抑制更新策略，並套用在三個經典得線上學習演算法。實驗結果顯示此框架在不同的觀念偏移資料下能有效提昇原先的學習演算法。Real-world online learning applications often face data coming from changing target functions. Such changes, called the concept drift, degrade the performance of traditional online learning algorithms. Thus, many existing works focus on detecting concept drift based on statistical evidence. Other works use sliding window or similar mechanisms to select the data that closely reflect current concept. Nevertheless, few works study how the detection and selection techniques can be combined to improve the learning performance. We propose a framework on top of existing online learning algorithms to improve the learning performance under concept drift. The framework detects the possible concept drift by checking whether forgetting some older data may be helpful, and then conduct forgetting through a step called unlearning. The framework effectively results in a dynamic sliding window that selects the data flexibly for different kinds of concept drifts. We design concrete approaches from the framework based on three popular online learning algorithms. Empirical results show that the framework consistently improves those online learning algorithms on ten synthetic data sets and two real-world data sets.496776 bytesapplication/pdf論文公開時間: 2016/8/25論文使用權限: 同意有償授權(權利金給回饋學校)線上學習觀念偏移資料Online learningconcpet driftthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/275541/1/ntu-104-R02922068-1.pdf