6.28 flush written test

Written examination total 22 topic , There are many types of questions , choice 、 Short answer 、 Programming 、 There are all open logic problems , The length of the written test 2h

There are many investigations on deep learning , Including performance evaluation indicators 、 Common definitions, etc

Multiple choice questions and short answer questions

1. Methods to improve the feature expression ability of deep neural network （ Which of the following methods can't be used to improve ） 

        a. Increase layer depth b. Use the nonlinear activation function c. Increase the number of hidden layer cells   Fine

2. Suppose the label of a clustering model is known , How to evaluate the performance of this model

3. explain batch-normalize The principle and implementation process of

4. Why introduce nonlinear activation function

5. use Hard Sample Mining What problems can technology solve

6. What is cross validation , Why use cross validation

7. In target detection , Recall rate 、 precision 、F1 How values are defined

        When I was doing it, I was thinking about the accuracy acc still pre, Finally, I feel that if it is pre Let's just say the accuracy , Yes acc It doesn't matter by definition

8. The direction along the negative gradient must be the optimal direction （ The statement that a multiple-choice question is wrong , I chose this ）

9. sketch sift Feature extraction method

10. Among the following algorithms, image smoothing is

        median filtering

11. Dense matrix A,B,C, Three The matrix dimensions are m*n,n*p,p*q, And m<n<p<q, The most efficient calculation sequence is

        （AB）C

12.CNN What are the weight initialization methods in

13. Explain the gradient descent 、 Stochastic gradient descent 、 The principle, advantages and disadvantages of small batch gradient descent

14. The following discussion about the time complexity of the algorithm , The correct statement is

        The time complexity of the algorithm has nothing to do with the selected programming language

15. In the neural network , Explain the activation function and softmax The difference between

16. To build a very deep network , Often used in convolution layer “vaild” Fill of , Only the pooling layer is used to reduce the width of the activation value 、 Height , Otherwise, the input will be reduced rapidly .（ Judge wrong or right ）

          wrong

17. It is known that the blood group distribution of Chinese is about A：30%,B：20%,O：40%,AB:10%, Select a group of Chinese people as the user research objects , I hope at least one of them is B The possibility of type a blood is no less than 90%, So at least how many people need to be chosen ？

        11 people   1-0.8^n>0.9

Open logic problems

Design a set of technical scheme for quantitative analysis of pressure ulcer         

        I'm a little confused when I see this question , Write something casually above and consider the various factors that induce pressure ulcers , Comprehensive evaluation

Programming questions

1. Find a continuous subarray , The sum of the major elements in the subarray and the maximum

        This question has been done to return the maximum sum , It's more complicated to return subarrays , I don't care so much when I do it ..

2. Find the target target The index of , If there is no , Then return to -1

        Try to catch up with the original question

3. Enter a linked list , Output the last number in the list K Nodes

        Try to catch up with the original question

4.（ Last question ） Write a super large positive integer class , Support addition operation

        First contact with this , mentally , Unclear java How to write , Help yourself to the python Write a few sentences

It seems that the whole set of papers has been reproduced , Make a note of , For future reference