当前位置：网站首页>Tencent offer has been taken. Don't miss the 99 algorithm high-frequency interview questions. 80% of them are lost in the algorithm

Tencent offer has been taken. Don't miss the 99 algorithm high-frequency interview questions. 80% of them are lost in the algorithm

2022-04-23 15:47:00 【InfoQ】

I've been 2021 In, he served as the algorithm group leader, As an interviewer, I interviewed many students . Interviewed more than 200 Famous students , Many of the students who joined the company developed well later , Also strengthened their self-confidence in the selection criteria .

This year, 2022 It's especially hard to find a job in , I sorted out some important interview questions as an interviewer over the years , altogether 80 Avenue , I hope I can help you .

For your convenience , I made a classification , It's divided into 6 Categories: , Namely ： machine learning , Feature Engineering , Deep learning ,NLP,CV, Recommendation system . This knowledge is a common question in an interview , It can also be used as a reference for everyone to sort out their own ideas .（ Students who need to get free at the end of the article ）

Machine learning theory ：

Write the total probability formula & Bayes' formula

Why introduce bias in model training (bias) And variance (variance)？ Prove

CRF/ Naive Bayes /EM/ Maximum entropy model / Markov random Airport / Gaussian mixture model

How to solve the over fitting problem ？

One-hot What is the role of ？ Why not just use numbers as a representation

What is the difference between decision tree and random forest ？

Naive Bayes why “ simple naive”？

kmeans The method of starting point other than random selection

LR It's clearly a classification model. Why is it called regression

How to parallelize gradient descent

LR Medium L1/L2 What is a regular term

Briefly describe the decision tree construction process

explain Gini coefficient

Advantages and disadvantages of decision tree

The estimated probability of occurrence is 0 How to deal with

The generation process of random forest

Introduce to you Boosting Thought

gbdt Of tree What is it? tree？ What are the characteristics

xgboost contrast gbdt/boosting Tree What are the optimization directions

What is an optimal hyperplane

What is support vector

SVM How to solve the multi classification problem

What is the function of kernel function

Characteristic Engineering ：

How to remove DataFrame The missing value in ？

Common operation methods of feature dimensionless

How to code class variables independently ？

How to make “ Age ” Fields are segmented according to our thresholds ？

How to draw a thermodynamic diagram according to the correlation of variables ？

How to modify the distribution to a normal like distribution ？

How to use PCA To partition the data and visualize it ？

How to use LDA To partition the data and visualize it ？

Deep learning ：

You feel batch-normalization What is the process like

What's the use of activating functions ？ What is the difference between common activation functions ？

Softmax What is the principle of ？ What's the role ？CNN What is the translation invariance of ？ How to achieve it ？

VGG,GoogleNet,ResNet What is the difference between such networks ？

Why can residual network solve the problem of gradient disappearance

LSTM Why can we solve the problem of gradient disappearance / The problem of explosion

Attention contrast RNN and CNN, What advantages do you think are

Write Attention Formula

Attention Mechanism , Inside q,k,v What do they stand for

Why? self-attention Can replace seq2seq

natural language processing （NLP） class ：

GolVe Loss function of

Why? GolVe Compare what you can use W2V Less

level softmax technological process

Negative sampling flow

How to measure what you learn embedding The stand or fall of

This paper CRF principle

detailed LDA principle

LDA How to calculate the topic matrix in

LDA and Word2Vec difference ？LDA and Doc2Vec difference

Bert Where is the two-way embodiment of

Bert How to pre train

Randomly select... From the data 15% The tag , among 80% Transposed [mask],10% unchanged 、10% Randomly replace other words , What's the reason

Why? BERT Yes 3 Two embedded layers , How they all come about

Writing a multi-head attention

Recommended system class ：

DNN And DeepFM The difference between

You're using deepFM How to deal with the problem of under fitting and over fitting

deepfm Of embedding Is there anything worth noting about initialization

YoutubeNet How to process variable length data

YouTubeNet How to avoid millions of softmax The problem of

What are the common evaluation indicators of the recommendation system ？

MLR What is the principle of ？ What optimizations have been made ？

Computer vision （CV） class ：

Common model acceleration methods

How to effectively solve the common problem of less foreground and more background in target detection

What's going on in target detection is SSD、YOLOv3、Faster R-CNN What can't be solved , Suppose the network fitting ability is infinitely strong

ROIPool and ROIAlign The difference between

Introduce common gradient descent optimization methods

Detection What else do you think you can do

mini-Batch SGD be relative to GD What are the advantages

What are the two mainstream methods of human posture estimation ？ A brief introduction

The realization principle of convolution and how to realize local convolution quickly and efficiently weight sharing Convolution operation of

CycleGAN Why is the generation effect of the general position unchanged texture changes , Why can't it produce generation effects in different positions
From this point of view, the algorithm is really important , So Xiaobian is here to share a book about algorithm Daniel

Specific branches of legal engineers ：

secondly , The necessary skills of Algorithm Engineers ：

▲ Familiar with at least one programming language C/C++/java/python/R;

▲ Skill level ; Proficient in using various common algorithms and data structures , Have independent implementation ability ;

▲ Familiar with data mining algorithms ;

▲ Familiar with machine learning related knowledge and theory .

▲ pluses ： Have rich experience in project practice .

Curious, you see here , There must be big questions ： Should we learn these algorithms directly ？

Ten thousand Zhang tall buildings rise from the ground , Any advanced algorithm should start from the basic algorithm , You can't be a fat man .

therefore , For beginners, you should start from the basics ：

▲ First, learn a language .

for example C/C++/Java/python, Beginner study C++ Quite common .

▲ Learn data structure .

There are many data structure books , But some textbooks are obscure , It is suggested to see more pictures , An easy to understand book , recommend 《 Interest in data structure 》.

▲ Learn algorithms .

Don't look directly 《 Introduction to algorithms 》, A lot of proof will break you down . recommend 《 Analysis of classical problems of data structure and algorithm 》, Problem analysis , Perfect illustration , Detailed explanation of dimension code , Practical drill , Suitable for beginners to quickly master classic algorithms .

Next , Let's follow 《 Analysis of classical problems of data structure and algorithm 》 Author's perspective , Find tips for learning algorithms and data structures ！

I don't want to talk much about it

If you are still in college, you can start with sorting and various basic data structures . I spent a week sorting and listing the eight Basics / Binary tree / Stack / The queue is made into a beautiful PDF.

This PDF Reading experience is definitely better than official account and blog posts .PDF The content is pure hand play ！

Here is a brief introduction to the eight basic sorting and basic data structure , The idea and basic explanation of each sort and the source code are in PDF Is punctuated with .