## Machine Learning ---- 一種讓機器自己找函式(function)的方式 ---- For example: Input: ![image](https://yt3.googleusercontent.com/6BCfAqi9yIpZbHLbw9BAWySvB3XZf9r8jFqudO5nSOsHoGzLhlKrm1M1uuMCRabi_pXGDzl7=s900-c-k-c0x00ffffff-no-rj =300x) $\rightarrow$ $f(x)$ = "Gura" Input: ![image](https://i.ytimg.com/vi/HvK28jxiz18/maxresdefault.jpg =300x) $\rightarrow$ $f(x)$ = defuse --- ## Types of functions ---- 回歸(Regression) 分類(Classification) ---- 回歸: 輸出值為數值 Tmr's raining possibility $\rightarrow$ $f(x)$ = 50% 2030's infant mortality rate $\rightarrow$ $f(x)$ = 3% ---- 分類: 輸出值為一類別(選擇題) Will tmr rain $\rightarrow$ $f(x)$ = yes/no ![image](https://hackmd.io/_uploads/rkJGgNggC.png =200x) $\rightarrow$ $f(x)$ = pdf/txt/jpg --- ## Generative Learning ---- Generative Learning(生成式學習) __=__ Structured Learning(結構化學習) ---- 找出一個input與output都是物件(object)的函式 生成出「有結構」的物件(Ex: 清單, 文句, 音樂) --- ## 找函式的架構 ---- A set of Functions(Model) $\downarrow$ Goodness of $f()$ $\uparrow$ Training Data --- ## 找函式的方法 ---- #### Supervised Learning #### Semi-supervised Learning #### Unsupervised Learning #### Transfer Learning #### Reinforcement Learning --- ### Supervised Learning 監督學習 把Training data給機器使其自行找到函式 ---- ### Training data 包含了function的input和output function的output別名為label,所以 __有給予function output__ 的training data亦稱為labelled data ---- __Classify Gura and Watson__ Training Data(Labels): ![image](https://hackmd.io/_uploads/SkbiwNlxA.png =70x) Watson ![image](https://hackmd.io/_uploads/ryHg_NxxC.png =70x) Gura ![image](https://hackmd.io/_uploads/SJe0vEee0.png =70x) Watson Training Data $\rightarrow$ __Model Training__ $\rightarrow$ $f()$ Test Data $\rightarrow$ $f(x)$ $\rightarrow$ Gura/Watson ---- ![image](https://hackmd.io/_uploads/Hyv9jNlxC.png) ---- Regression, Classification, Structured learning都是Supervised learning的方法 --- ### Semi-supervised Learning 半監督學習 ---- 結合了labelled data與unlabelled data給機器尋找函式,不過這裡的unlabelled data有一個綜合的function output ---- ### Unlabelled data 與labelled data不同,這裡的training data並沒有經過label --- ### Unsupervised Learning無監督學習 僅使用unlabelled data給機器訓練並找出函式 預測資料之間的關係 (即無function output) ---- Unlabelled data--> use clustering ---- Example: 給予大量文章(unlabelled data)並讓機器產生某個單詞的解釋 ![image](https://hackmd.io/_uploads/rJn76gGuR.png) --- ### Transfer Learning 轉移學習 預先訓練一個模型(model)並把他轉移至另一個模型 減少額外訓練時間 --- ### Reinforcement Learning 強化學習 只有function input,給function output打分數 ---- Reward: 0 3+1 $\rightarrow$ f(x) $\rightarrow$ 2 Reward: 1 3+1 $\rightarrow$ f(x) $\rightarrow$ 4
{"title":"ML Intro","description":"Reference: \n李弘毅 videos","contributors":"[{\"id\":\"d4f42d78-aa45-44e2-ade2-4d4e233d6166\",\"add\":4173,\"del\":1570}]"}
    118 views