AI演算法面試考題整理

整理電腦文件的時候發現之前留下來的AI演算法面試題目，所幸彙整一下順便做個紀錄。

以下題目節錄自面試過的幾個AI演算法相關職缺時遇到的考題，其中主要包含ML/DL演算法、訊號處理、過往訓練模型經驗等。由於有些公司以全英文出題，有的則是中文考題，因此本文附上中英文題目，供參考。

ML/DL算法

1. 什麼是Support Vector Machine (SVM)? 請詳細解釋。
   What is Support Vector Machine (SVM)? Please explain SVM in detail.
2. 解釋CNN、RNN差別，Resnet原理以及好處。
   Explain the differences between CNN and RNN, as well as the principles and benefits of ResNet.
3. 解釋cross-entropy。
   Explain cross-entropy.
4. 推導Backpropagation。（題目會附上一張back propagation的圖，請你推導如何更新某個節點的weight。）
   Backpropagation derivation. (The question would provide a diagram of backpropagation. Please derive how to update the weights, such as w¹¹, w¹², w²¹…)
5. 訓練模型時為什麼通常需要正則化？請解釋L1和L2正則化的差異。
   Why regularization is usually necessary when training a model? Explain the difference between L1 and L2 regularization.
6. 什麼是 confusion matrix？ 怎麼從 confusion matrix 計算 sensitivity 以及 precision？
   Exaplain confusion matrix. How to calculate sensitivity and precision with a confusion matrix? Explain sensitivity and precision.
7. 請描述偏差（Bias）與變異（Variance）之差別，訓練過程哪項指標更重要？
   Please describe the difference between bias and variance, and which one is more important during the training process.
8. 要如何做 Hyper-parameter tuning，請說明方法及原因。
   How to do Hyper-parameter tuning? Please explain the method and reason.
9. KNN和multi-nominal logistic regression之間有什麼不同？
   What are the differences between K nearest neighbor ( KNN ) and multi-nominal logistic regression?
10. 何時需要進行降維？請詳細描述兩種降維方法。
    When to do dimension reduction? Please describe 2 methods of dimension reduction in detail.
11. 除了Mean Square Error (MSE)之外，如何評估回歸模型？
    Except for mean square error, how to evaluate the quality of a regression model?

訊號處理

1. 解釋傅利葉中FS, FT, DTFT, DFT 以及FFT差別。
   Explain the differences between FS, FT, DTFT, DFT, and FFT.
2. 對於參雜未知雜訊之訊號，如何評估訊雜比？
   How to evaluate the signal-to-noise ratio (SNR) for a signal contaminated with unknown noise?
3. 如何對無限訊號應用低通濾波器？ 請詳細描述訊號處理流程。
   How to apply a low pass filter to an infinite signal? Please describe the signal processing flow in detail.
4. 如何找到或偵測訊號中的某種模式？
   How to find or detect a certain pattern in a signal?
5. Convolution和Correlation有什麼不同？什麼狀況下會他們會相同？
   What are the differences between convolution and correlation? When are they the same?

其他

1. 如果資料類別極度不平衡，建立模型後在測試集依然達到了99%的準確度（Accuracy），這會造成什麼問題？
   If the data are extremely unbalanced and a model achieves 99% accuracy on the testing data, what problems could this cause?
2. 假設可以拿到無限的資料，你需要哪一些資料來幫助你做訊號、影像分類？你的分類流程是什麼？
   Assume you could obtain any data you need, what data do you need to aid in signal or image classification? What is your classification process?
3. 請列舉一項做過最熟的理論或演算法並解釋。
   Please list and explain one theory concept or algorithm you are most familiar with.
4. 除了Machine learning的方法，你是否熟悉其他傳統影像處理演算法？請舉例。
   In addition to machine learning methods, are you familiar with other traditional image processing algorithms? Please provide examples.

因為是幾個月前的面試了，有些非考卷的問答題忘記了，因此只列舉了上述這些🥲 希望大家都能在寒冬上岸🥹