4. Which of the following statements about artificial neural networks is wrong?以下哪一個關於人工神經網路的描述是錯誤的？Back propagation is an important feature of neural networks: information from the output layer will be fed back to the hidden layers and the input layer to further improve the performance of the model. 反向傳播是神經網路的一個重要特徵：輸出層會將資料反向傳送囘隱藏層和輸入層，從而進一步改善神經網路的表現。There is always at least one output layer. 一定至少有一個輸出層。In a fully connected layer, all neurons are connected to all the other neurons in the previous layer and those in the next layer. 在全連接層中，所有神經元都會連接到上一層以及下一層中的所有其他神經元。There are always more input layers than hidden layers. 輸入層一定比隱藏層多。

以下哪一個關於人工神經網路的描述是錯誤的？There is always at least one output layer. 一定至少有一個輸出層。In a fully connected layer, all neurons are connected to all the other neurons in the previous layer and those in the next layer. 在全連接層中，所有神經元都會連接到上一層以及下一層中的所有其他神經元。There are always more input layers than hidden layers. 輸入層一定比隱藏層多。Back propagation is an important feature of neural networks: information from the output layer will be fed back to the hidden layers and the input layer to further improve the performance of the model. 反向傳播是神經網路的一個重要特徵：輸出層會將資料反向傳送囘隱藏層和輸入層，從而進一步改善神經網路的表現。

4. Which of the following is true?以下哪一個説法正確？‘Deep Learning’ is a subfield of ‘Machine Learning’. 「深度學習」涵蓋於「機器學習」當中。‘Machine Learning’ is also known as ‘Deep Learning’. 「機器學習」是「深度學習」的另一種叫法。There is no relation among ‘AI’, ‘Machine Learning’ and ‘Deep Learning’.「人工智能」、「機器學習」和「深度學習」是三種沒有關係的概念。‘Machine Learning’ is a subfield of ‘Deep Learning’. 「機器學習」涵蓋於「深度學習」當中。

5. How does a single-layer artificial neural network handle input values it receives?一個單層的人工神經網路如何處理接收到的輸入值？ Multiply with weights → Sum together → Pass to activation function → Output 乘以權重 → 加總 → 傳入激勵函數 → 輸出Pass to activation function → Multiply with weights → Sum together → Output 傳入激勵函數 → 乘以權重 → 加總 → 輸出Sum together → Pass to activation function → Multiply with weights → Output 加總 → 傳入激勵函數 → 乘以權重 → 輸出Multiply with weights → Pass to activation function → Sum together → Output 乘以權重 → 傳入激勵函數→ 加總 → 輸出

6. Which of the following is true?以下哪一個説法正確？The inferred results generated by Machine Learning are always correct. 機器學習推斷出的結果總是正確的。Machine Leaning is able to infer an absolutely correct result in some situations, while in some other situations, it can only provide probabilities of different possible options. 機器學習在某些情況下能夠作出絕對正確的推斷，某些情況卻只能計算不同選項的機率。Machine Learning is not able to make inference and provide probabilities of different possible options. 機器學習既無法作出推斷，也無法給出不同可能選項的機率。The inferred results generated by Machine Learning are usually accompanied with probabilities, which indicate the confidence of the inference. 機器學習除了提供推斷結果之外，通常還會附上該推斷結果發生的機率。這個機率表示對該推斷結果的信心。

Please select the best and correct answer. 請選出最好且正確的答案。1. ‘Teachable Machine’ is a machine learning tool discussed in our courses. It provides three modes, Image Project, Audio Project and Pose Project, to complete an AI project. Which of the following statements about the operation of ‘Teachable Machine’ is wrong?我們的課程中討論了一個機器學習工具 ‘Teachable Machine’。此工具提供使用者圖像辨識、聲音辨識、肢體辨識三種模式以完成人工智能專案。以下哪個對 ‘Teachable Machine’運作的說法是錯誤的？The steps of ‘training’ include calculating the optimal weights of an artificial neural network.「訓練」的步驟包括計算出最佳化的神經元傳輸權重值。Inputting more data of good quality helps to improve the identification rate. 匯入品質好的數據可幫助改善辨識率。‘Teachable Machine’ belongs to the field of unsupervised learning, as it does not require any labelled data to let the computer categorize. ‘Teachable Machine’屬於非監督式學習領域，因為過程中不需要使用有標籤的資料讓電腦自行分類。The function of adding ‘class’ is to label the imported data, so as to teach the computer to learn how to categorize according to the labels. 輸入「分類」就是讓我們標籤輸入的數據，以便教電腦學習哪一類的樣本數據是屬於哪一種標籤。2. Carmen has been collecting data, and she hopes to use AI to study the relationship between interviewees’ income level and various factors such as education level, age and gender. Later, she finds that the interview dates are missing for 80% of the interviewee data. Which of the following is the most reasonable action for her to take before training her AI model?嘉雯一直在收集數據，希望利用人工智能找出受訪者收入水平與其他因素（例如教育水平、年齡、性別）的關係。後來，她發現有八成受訪者的數據缺失了訪問日期。以下哪個是她在訓練她的人工智能模型之前最合理的行動？Standardizing the date format 把日期格式統一Substituting the missing entries with a random date 在缺失日期的欄目補上一個隨機的日期Discarding the column for interview dates in all the interviewee data 捨去全部受訪者數據中的訪問日期一欄Discarding the individual interviewee data with missing interview dates 捨去缺失了訪問日期的個別受訪者數據