![](https://i.imgur.com/mYvboeO.png) # 林妤敏 工作日誌         07月01號 第1週 ## 第一週 ### Artificial Inteligence : Machine Learning **學習方式** | Type | Supervised Learning | Unsupervised Learning | Reinforcement Learning | | ---------- | --------------------- | ----------------------- | ---------------------- | | Algorithms | Linear/Logistic Reg., | Kmeans,DBSCAN,PCA,NN, | Q-Learning, SARSA | | | SVM,Decision Tree,KNN | Hierarchical Clustering | | | Train | labeled data | unlabeled data | No predefined data | | | need supervised | no need supervised | self-learning | | Type of | Classification | Clustering, Apriori | Reward based system | | Problems | Regression | Association | | | | | Receonstruct | | | Examples | Email,Forecast | 購物平臺的自動推薦功能 | Gaming | **4 type of Classification** * Binary Classification 二元分類 : Email Spam(Yes or No) * Multi-Class Classification 多元分類 : Face recognition system * Multi-Label Classification 多標簽分類: Photo Classification * Imbalanced Classification 不平均分佈 : Medical Diagnostic system **Regression** * 透過很多的**x**去預測出**y**這個結果 * 預測的答案為連續值 * 輸出結果是數值 **總結訓練過程** 【訓練前】 1. Scenario :有什麽資料?要預測什麽資料? 2. Task :問題種類:要做什麽?解什麽? 3. Method: 要用什麽方法?決定Model(A set of function) 【訓練中】 1. Loss Function損失函數: train他如何辦別模型的好壞(Goodness of function f) 2. Gradient Descent梯度下降法: 到了最佳的 (Pick the Best function) 3. Learning Rate: 可以調整學習脚步 【訓練後】 1. Test Avearage Error訓練可能的問題 2. Generalization: 評估新資料的適用性 3. Cross Validation: Testing Set驗證Model的好壞 **error的來源** bias & variance ![](https://i.imgur.com/Uoqk5M5.jpg) ![](https://i.imgur.com/hPKY1QI.jpg) ![](https://i.imgur.com/C6vwEuq.png) 比較簡單的圖:Bias大,Variance小 比較複雜的圖:Bias小,Variance大 Overfitting (模式複雜度過高): Bias2降低,但Variance增加 Underfitting (模式複雜度過低): Variance降低,但Bias2增加 ## 下週預期進度 Deep Learning TCA course AI Conculsion