(1) Introduce yourself.
(2) Introduce the project.
Interviewer: You have been working on watermark removal recently. Can you introduce this project?
Me: I talked to the interviewer about my watermarking work, some papers I read, and my future work on this project.
(3) Check the clustering algorithm.
Interviewer: Can you briefly describe what clustering is?
Me: I talked about the different algorithms of clustering: prototype clustering, density clustering, and hierarchical clustering. Prototype clustering includes: K-means algorithm, Gaussian mixture clustering, and one forgot, which is this (learning vector quantization).
Interviewer: Can you introduce density clustering?
Me: I can't remember this! I haven't watched it for a long time, I forgot.
(4) Evaluation regression algorithm.
Interviewer: Briefly introduce the regression algorithm.
Me: I gave an introduction to the regression algorithm.
Interviewer: What are the commonly used regression algorithms?
Me: Regression algorithms are not often used, I understand linear regression. I explained linear regression to him.
Interviewer: Do you know about Ling's return?
Me: Ridge regression should be used to solve model overfitting! It should be L1 regular or L2 regular!
(5) Assess L1 regularity and L2 regularity
Interviewer: Then you can talk about L1 regularity and L2 regularity!
Me: I told him about the L1 regular formula and the L2 regular formula. Also, why do L1 and L2 regularities appear, and why L1 regularities and L2 regularities solve the problem of overfitting.
(6) The problem of matrix decomposition.
Interviewer: Do you know how matrix decomposition is used?
Me: SVD, Singular Value Decomposition. LDA topic model, I don't quite understand this one! I don't know if it can solve the problem of matrix decomposition! (Speaking of this, I feel very underpowered!)
Interviewer: Then can you talk about how SVD does matrix decomposition?
Here. . . Haven't done it! SVD is also largely forgotten. . . . . . . . (sad....)
(7) Deep Learning (Recurrent Neural Network)
Interviewer: Looking at your resume, you know about the recurrent neural network of deep learning. Have you ever done any projects in this area?
Me: I've written papers on this, doing stock forecasting. I listed the shortcomings of the recurrent neural network: gradient disappearance and gradient explosion, etc., and then talked about why gradient disappearance and gradient explosion occurred. Finally, why did I choose LSTM and the principle of LSTM, and how to overcome gradient disappearance and gradient. explosive. and how to use it in my dissertation. . Hold on!
(8) Deep Learning (Convolutional Neural Network)
Interviewer: Tell me about Convolutional Neural Networks!
Me: I just told him about convolutional neural networks. Didn't ask too specific.
(9) Project details.
Interviewer: You said before that you do watermark removal, so how do you detect the position of the watermark?
Me: I told him the algorithm for detecting watermarks.
(10) Deep Learning (Optimizing Neural Networks)
Interviewer: In deep learning, how to optimize neural network?
我：dropout、Normalization（Batch Normalization、weight Normalization、Layer Normalization、Cosine Normalization）等。
Interviewer: Then do you know the difference between Layer Normalization and Batch Normalization?
Me: Not very clear!
Interviewer: Then tell me the principle of dropout!
Me: I told him the principle of dropout. (I feel like I'm not good at speaking! Some details are not very clear! Take the time to review it recently!)
(11) FINAL QUESTIONS
Interviewer: Is there anything you want to ask me at the end?
Me: If this side of CVTE is over, how many sides are there?
Face Master: One or both sides!
Me: Do you think there are any flaws in my knowledge system and knowledge structure? Do you have any suggestions for my current body of knowledge and knowledge architecture?
Interviewer: Everyone's knowledge system and knowledge structure are different, so it's hard to give advice. But doing machine learning and deep learning, project experience is still very important. Let's strengthen the project! We still hope that everyone has a different knowledge system and knowledge structure, so that a team will be better. Probably so!
The interview time is about 27 minutes.
I don't know if I can make it through! Hope you can make it through!