English  |  正體中文  |  简体中文  |  全文筆數/總筆數 : 80990/80990 (100%)
造訪人次 : 43368855      線上人數 : 1363
RC Version 7.0 © Powered By DSPACE, MIT. Enhanced by NTU Library IR team.
搜尋範圍 查詢小技巧:
  • 您可在西文檢索詞彙前後加上"雙引號",以獲取較精準的檢索結果
  • 若欲以作者姓名搜尋,建議至進階搜尋限定作者欄位,可獲得較完整資料
  • 進階搜尋


    請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/86587


    題名: An Efficient Cluster-Based Continual Learning with Gradient Episodic Cache Memory
    作者: 陳鈺欣;Chen, Yu-Hsin
    貢獻者: 資訊管理學系
    關鍵詞: 機器學習;深度學習;連續學習;聚類分析;Machine Learning;Deep Learning;Continual Learning;Cluster Analyze
    日期: 2021-07-20
    上傳時間: 2021-12-07 13:00:15 (UTC+8)
    出版者: 國立中央大學
    摘要: 近年來,深度學習已變得越來越流行,它已被廣泛應用於各個領域並取得了優異的成績,但是深度學習通常在很少的訓練樣本的情況下無法達到預期的結果,並且它應該像人類一樣能夠利用過去的經驗來快速學習新任務,因此持續學習的重要性明顯增加,而主要目標是在不忘記過去所學知識的情況下學習新任務。首先,我們提出了一種名為Gradient Episodic Cache Memory的方法,結合了聚類技術來解決Gradient Episodic Memory的存儲和計算問題。其次,我們在CIFAR-10、CIFAR-100和MNIST Permutations數據集上評估模型,而實驗結果表明,GECM的性能優於其他的連續學習的模型,並且GECM在準確性和效率之間也取得了良好的平衡。;In recent years, deep learning has become more and more popular. It has been widely used in various fields and has achieved outstanding results. However, deep learning usually fails to achieve the expected results in the condition of few training samples and it should be able to use past experience to quickly learn new tasks like human beings. Therefore, the importance of continuous learning increases significantly while the main goal is to learn new tasks without forgetting what has been learned in the past. First, we propose our method called Gradient Episodic Cache Memory (GECM), which is based on Gradient Episodic Memory framework and combines clustering techniques to resolve the memory and computation problems of Gradient Episodic Memory. Second, we evaluate our model on CIFAR-10, CIFAR-100, and MNIST permutations datasets. The experimental results show that GECM performs better than other state-of-the-art continual models and GECM has a good balance between accuracy and efficiency.
    顯示於類別:[資訊管理研究所] 博碩士論文

    文件中的檔案:

    檔案 描述 大小格式瀏覽次數
    index.html0KbHTML60檢視/開啟


    在NCUIR中所有的資料項目都受到原著作權保護.

    社群 sharing

    ::: Copyright National Central University. | 國立中央大學圖書館版權所有 | 收藏本站 | 設為首頁 | 最佳瀏覽畫面: 1024*768 | 建站日期:8-24-2009 :::
    DSpace Software Copyright © 2002-2004  MIT &  Hewlett-Packard  /   Enhanced by   NTU Library IR team Copyright ©   - 隱私權政策聲明