機器學習及其應用2023

目 錄

“生成一切”背後的數學原理 ······································································雷 娜顧險峰  1 1 傳統圖像處理方法 ···································································································· 1 2 圖像生成算法 ··········································································································· 2 3 3D曲面生成算法 ······································································································ 6 4 未來展望··················································································································· 8 參考文獻························································································································· 9 

高維樣本協方差矩陣的譜性質及其應用簡介 ···························王瀟逸鄭術蓉鄒婷婷  11 1 引言··························································································································11 2 高維框架下傳統方法失效的例子 ·········································································· 12 3 大維樣本協方差矩陣的極限譜分佈 ······································································ 14 4 大維樣本協方差矩陣的應用 ·················································································· 17 5 總結和展望 ············································································································· 25 參考文獻······················································································································· 25 

多目標演化學習：理論與算法進展 ··········································································錢超 27 1 引言························································································································· 27 2 理論分析工具 ——調換分析·················································································· 29 3 理論透視················································································································· 33 4 多目標演化學習算法 ······························································································ 38 5 總結與展望 ············································································································· 45 參考文獻······················································································································· 46 

自監督學習的若乾研究進展 ······················································楊健陳碩李翔 49 1 引言························································································································· 49 2 相關工作················································································································· 51 

機器學習及其應用 2023 

3 基於對比學習與自編碼學習的自監督學習算法 ··················································· 54 4 總結與展望 ············································································································· 74 參考文獻······················································································································· 74 

因果性學習 ··················································································李梓健蔡瑞初郝志峰 78 1 引言························································································································· 78 2 基於先驗因果結構的因果性學習方法及其應用 ··················································· 80 3 基於因果發現的因果性學習方法及其應用 ··························································· 87 4 小結························································································································· 91 參考文獻······················································································································· 92 

先排序後微調：預訓練模型庫利用的新範式 

······························游凱超劉雍張子陽王建民  Michael I. Jordan 龍明盛 95 1 引言························································································································· 95 2 相關工作················································································································· 98 3 對預訓練模型進行排序 ·························································································102 4 LogME算法的理論分析························································································108 5 預訓練模型微調 ·····································································································112 6 實驗························································································································116 7 結論························································································································131 附錄··································································································································· 132 A 符號對照表············································································································132 B 定理 1證明············································································································133 C 定理 2證明············································································································134 D 推論 1證明············································································································135 E 推論 2證明············································································································137 F 數據集描述 ············································································································139 G 圖表的原始結果 ····································································································139 H 提示學習完整結果 ································································································142 I 收斂性分析完整圖表 ·····························································································142 參考文獻······················································································································145 

目 錄 

遷移學習 ···················································································································莊福振 150 

1 引言························································································································150 2 相關工作················································································································152 3 概述························································································································153 4 基於數據的解釋 ·····································································································156 5 基於模型的解釋 ·····································································································173 6 應用························································································································185 7 實驗························································································································189 8 結論和未來方向 ·····································································································195 參考文獻······················································································································196 

基於表示學習的機器學習模型復用 ·······································································葉翰嘉 211 1 引言························································································································211 2 模型復用背景 ········································································································213 3 模型復用方法 ········································································································215 4 可復用模型方法 ·····································································································226 5 總結與展望 ············································································································238 參考文獻······················································································································239 

並行算法組自動學習研究簡介 ································································劉晟材唐珂 241 1 引言························································································································241 2 相關工作················································································································243 3 並行算法組自動學習 ·····························································································245 4 總結························································································································262 參考文獻······················································································································263 

ChatGPT的演進歷程與未來發展趨勢 ···················································朱慶福車萬翔 265 1 引言························································································································265 2 相關工作················································································································266 3 ChatGPT概覽 ········································································································267 4 ChatGPT的關鍵技術與解決的關鍵科學問題······················································269 5 ChatGPT對自然語言處理的影響·········································································271 6 總結和展望 ············································································································275 參考文獻······················································································································275