十大值得關注的深度學習算法

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"預測未來不是魔法，而是人工智能。毋庸置疑，人工智能的風頭正勁，每個人都在談論它，無論他們是否理解這個術語。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"據研究人員和分析師稱，到 2024 年，數字助理的使用率預計有望達到 "},{"type":"link","attrs":{"href":"https:\/\/www.semrush.com\/blog\/artificial-intelligence-stats\/","title":null,"type":null},"content":[{"type":"text","text":"84 億"}],"marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"。超個性化、聊天機器人、預測性行爲分析等是人工智能應用中最爲常見的用例。人工智能正在徹底改變整個地球，引領我們走向一個不可預測的未來。其中，最重要的兩個概念是機器學習和深度學習。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"在每天發送的 3000 億封電子郵件中，"},{"type":"link","attrs":{"href":"https:\/\/dzone.com\/articles\/a-beginners-guide-to-machine-learning-what-aspirin","title":null,"type":null},"content":[{"type":"text","text":"機器學習"}],"marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"的效率足以檢測出垃圾郵件。但是，近年來，深度學習以其高準確率、有效性、高效率和處理海量數據的能力而受到廣泛歡迎。它是機器學習的一個分支，通過學習將整個世界呈現爲一個根深蒂固的概念層次，每個概念都被確定爲簡單，從而擁有巨大的靈活性和力量。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"隨着人工神經網絡的應用，深度學習算法訓練機器在大量數據上進行復雜的計算。深度學習算法可以讓機器能夠像人腦那樣進行工作和處理數據，並高度依賴於人工神經網絡，並基於人腦的結構-功能而工作。以下是十大值得關注的深度學習算法，希望能對你有所參考。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"1. 自動編碼器"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"作爲某種類型的前饋神經網絡，"},{"type":"link","attrs":{"href":"https:\/\/towardsdatascience.com\/applied-deep-learning-part-3-autoencoders-1c083af4d798","title":null,"type":null},"content":[{"type":"text","text":"自動編碼器"}],"marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"（Autoencoder）是一種深度學習算法，其中輸入和輸出都是相同的。它是由 Geoffrey Hinton 在 1980 年設計的，目的是解決無監督學習問題。它擁有經過訓練的神經網絡，將數據從輸入層轉移到輸出層。自動編碼器的一些重要用例是：圖像處理、藥品回收和人口預測。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"以下是自動編碼器的三個主要組成部分："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"編碼器（encoder）"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"編碼（Code）"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"解碼器（decoder）"}]}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"2. 受限玻爾茲曼機"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https:\/\/wiki.pathmind.com\/restricted-boltzmann-machine","title":null,"type":null},"content":[{"type":"text","text":"受限玻爾茲曼機"}],"marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"（Restricted Boltzmann Machines，RBM）是隨機神經網絡，能夠從概率分佈而不是輸入集中學習。這種深度學習算法由 Geoffrey Hinton 開發，用於主題建模、特徵學習、協同過濾、迴歸、分類和降維。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"受限玻爾茲曼機分兩個階段工作："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"前向（forward pass）"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"後向（backward pass）"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"此外，它由兩個層組成："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"隱含層（Hidden units）"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"可視層（Visible units）"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"每個可見層都與所有現有的隱含層相連。受限玻爾茲曼機也有一個偏置層。該層與所有的隱含層以及可見層相連，但沒有輸出節點。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"3. 自組織映射"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https:\/\/dzone.com\/articles\/self-organizing-maps","title":null,"type":null},"content":[{"type":"text","text":"自組織映射"}],"marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"（Self-Organizing Maps，SOM）通過自組織人工神經網絡實現了數據可視化，以降低數據的維度。這種深度學習算法是由 Teuvo Kohonen 教授開發的。數據可視化能夠解決人類在處理高維數據時不容易可視化的這類問題。開發自組織映射的目的是爲了對高維信息進行更好的理解。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"4. 多層感知機"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"開始學習深度學習算法的最好地方是"},{"type":"link","attrs":{"href":"https:\/\/www.sciencedirect.com\/topics\/computer-science\/multilayer-perceptron","title":null,"type":null},"content":[{"type":"text","text":"多層感知機"}],"marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"（Multilayer Perceptions，MLP）。它屬於前饋神經網絡的範疇，同時還有許多包含激活函數的感知層。 它由兩個完全連接的層組成："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"輸入層"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"輸出層"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"多層感知機包含相同數量的輸入層和輸出層，並且有可能具有各種隱含層。多層感知機的一些重要用例包括圖像識別、人臉識別和機器翻譯軟件。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"5. 深度信念網絡"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"生成模型、"},{"type":"link","attrs":{"href":"https:\/\/medium.com\/@icecreamlabs\/deep-belief-networks-all-you-need-to-know-68aa9a71cc53","title":null,"type":null},"content":[{"type":"text","text":"深度信念網絡"}],"marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"（Deep Belief Network，DBN）擁有大量的潛變量和隨機變量層。潛變量通常被稱爲隱含層，包含二進制值。這些是波爾茲曼機的堆棧，各層之間有連接。每一個深度信念網絡層都與後續和之前的層相連接。深度信念網絡的用例包括視頻識別、圖像識別以及運動捕捉數據。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"6. 徑向基函數網絡"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https:\/\/www.sciencedirect.com\/topics\/chemical-engineering\/radial-basis-function-networks","title":null,"type":null},"content":[{"type":"text","text":"徑向基函數網絡"}],"marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"（Radial Basis Function Network ，RBFN）是一類特殊的前饋神經網絡，利用徑向基函數作爲激活函數。它包含以下幾層："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"輸入層"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"隱含層"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"輸出層"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"上述各層的徑向基函數網絡被用來進行迴歸、分類和時間序列預測。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"7. 生成對抗網絡"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https:\/\/machinelearningmastery.com\/what-are-generative-adversarial-networks-gans\/","title":null,"type":null},"content":[{"type":"text","text":"生成對抗網絡"}],"marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"（Generative Adversarial Network，GAN）是一種深度學習算法，它可以創建與訓練數據相似的新數據實例。生成式對抗網絡有助於生成逼真的圖片、卡通人物、人臉的圖像創建和三維物體的渲染。視頻遊戲開發者利用生成對抗網絡，通過圖像訓練提升低分辨率。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"生成對抗網絡有兩個重要組成部分："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"生成器（generator）：能夠生成虛假數據。"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"鑑別器（discriminator）：能夠從虛假信息中學習。"}]}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"8. 遞歸神經網絡"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https:\/\/stanford.edu\/~shervine\/teaching\/cs-230\/cheatsheet-recurrent-neural-networks","title":null,"type":null},"content":[{"type":"text","text":"遞歸神經網絡"}],"marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"（Recurrent Neural Network，RNN）由有助於形成有向循環的連接組成，允許長短期記憶網絡（Long Short-term Memory Network，LSTM）的輸出作爲現階段的輸入提供。遞歸神經網絡能夠記住以前的輸入，因爲它有內部記憶。遞歸神經網絡的一些常見用例有：手寫識別、機器翻譯、自然語言處理、時間序列分析和圖像說明。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"9. 卷積神經網絡"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https:\/\/en.wikipedia.org\/wiki\/Convolutional_neural_network","title":null,"type":null},"content":[{"type":"text","text":"卷積神經網絡"}],"marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"（Convolutional Neural Network，CNN）也被稱爲 ConvoNet，包含許多層，主要用於物體檢測和圖像處理。第一個卷積神經網絡是由 Yann LeCun 在 1988 年開發和部署的。在那一年，它被稱爲 LeNet，用於字符識別，如數字、郵政編碼等。卷積神經網絡的一些重要用例包括醫學圖像處理、衛星圖像識別、時間序列預測和異常檢測。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"以下是卷積神經網絡的一些關鍵層，它們在數據處理以及從數據中提取特徵方面起着舉足輕重的作用："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"卷積層"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"線性整流層"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"池化層"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"完全連接層"}]}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"10. 長短期記憶網絡"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https:\/\/machinelearningmastery.com\/gentle-introduction-long-short-term-memory-networks-experts\/#:~:text=Long%20Short%2DTerm%20Memory%20(LSTM,complex%20area%20of%20deep%20learning.","title":null,"type":null},"content":[{"type":"text","text":"長短期記憶網絡"}],"marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"（Long Short-term Memory Network，LSTM）是一類遞歸神經網絡，能夠學習和記憶長期依賴關係。長短期記憶網絡還能夠長期回憶過去的信息。它能隨着時間的推移保留信息，這被證明在時間序列預測中是有益的。它有一個鏈狀結構，其中 4 個相互作用的層連接並進行獨特的溝通。除了時間序列預測外，長短期記憶網絡還被用於藥品開發、音樂創作和語音識別。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"總結"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"近年來，深度學習算法和技術之所以流行，主要原因在於它們能夠處理大量的數據，然後將數據轉化成信息。利用它的隱含層架構，深度學習技術學會了定義低級別的類別，如字母；然後是中級別的類別，如單詞；然後是高級別的類別，如句子。據某些預測，深度學習必將徹底改變供應鏈自動化。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"百度前首席科學家，以及谷歌大腦項目的著名領導人之一，"},{"type":"link","attrs":{"href":"https:\/\/www.wired.com\/brandlab\/2015\/05\/andrew-ng-deep-learning-mandate-humans-not-just-machines\/","title":null,"type":null},"content":[{"type":"text","text":"吳恩達"}],"marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}]},{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"（Andrew Ng）曾肯定地表示："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"與深度學習類似的是，火箭發動機是深度學習模型，燃料是我們可以提供給這些算法的海量數據。（"},{"type":"text","marks":[{"type":"italic"}],"text":"“The analogy to deep learning is that the deep learning models are the rocket engines and the immense amount of data is the fuel to those rocket engines. ”"},{"type":"text","text":"）"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"因此，技術的發展和進步永遠不會停止，深度學習技術和算法也是如此。爲了在這個日新月異的世界中保持競爭力，每個人都必須跟上最新的科技進步。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}},{"type":"strong"}],"text":"作者介紹："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"Aliha Tanveer，技術作家，供職於 ArhamSoft。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}},{"type":"strong"}],"text":"原文鏈接："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"color","attrs":{"color":"#494949","name":"user"}}],"text":"https:\/\/dzone.com\/articles\/10-crucial-deep-learning-algorithms-to-keep-an-eye"}]}]}