At the end of the article Easter egg: July online dry goods group’s latest upgrade of “AI interview 100 famous enterprises” free to send!
01 Personal Information
I am not a computer major, after graduation, coincidence into the algorithm position, with July is the computer major seniors recommended the question bank in July.
I was mainly exposed to traditional machine learning algorithm in my previous work, and I only had a superficial understanding of NLP. I found that traditional machine learning algorithm was not enough in the interview, so I applied for the NLP employment class in July.
02 Learning Situation
In the preparatory stage, I went through the traditional machine learning algorithm, but I didn’t have time to pass the deep learning and NLP preparatory courses, which directly led to my first homework being completely dry. It took me a week, and I couldn’t bear to look at what I handed in (I couldn’t imagine how Dr. Chu could bear to watch it).
After realizing my taste, I started taking NLP classes, along with deep learning, PyTorch and Tensorflow. The most afraid of miss Chu’s live class, because I didn’t hand in my homework, didn’t finish the exercises, didn’t finish the code, so every time I saw the teacher shivered.
Although I am afraid of teachers, I always think THAT I am very lucky. At the beginning of my algorithm road, I can have a professional teacher to guide me, which saves a lot of time for myself to explore.
Dr. Chu showed us English papers, analyzed source code, taught us the latest TECHNOLOGY of NLP, provided rich learning resources, suggested the modification of resume, optimized project and so on. During this process, WHAT I learned was not only the current technology of NLP field, but also a change of thinking mode and learning method. A qualified algorithm engineer should have professional quality and personal cultivation, which is far more important than the formula derivation and application of a model.
In the IT industry, the update and iteration cycle is getting shorter and shorter, and certain models will soon be eliminated. However, excellent ways of thinking, learning methods and self-cultivation are always very competitive. I think this is the biggest harvest I gained from following Dr. Chu in July.
03 Interview Situation
The interview process is generally divided into the following stages: self-introduction, project introduction and questions, code and algorithm assessment. Self-introduction, project introduction, algorithm assessment is necessary regardless of the size of the company, data structure part of the company assessment.
Data structure:
(1) Tear quick sort, bubble sort, merge sort;
Algorithm:
(1) Introduction of the principle and steps of logistic regression, loss function of handwritten logistic regression; Why is serial multiplication in the loss function of logistic regression? Can I add?
(2) Why do we need to do maximum likelihood estimation?
(3) The realization of k-means principles and steps (oral introduction and writing on the board), and the code of K-means by hand; Limitations and optimization methods; What other methods are there for clustering besides K-means?
(4) How many steps can EM algorithm be divided into? What are the steps? What are the differences and connections between EM algorithm and K-means algorithm?
(5) Differences and applicable places of ID3, C4.5 and CART;
(6) The principle of SVM and CRF algorithm; Whether the optimal value obtained by SVM is global optimal; How do you determine whether an optimal value is globally optimal or locally optimal?
(7) The difference between HMM and CRF;
(8) What are the classification algorithms in the multidimensional space?
(9) There is a circle, one point is inside the circle, and one point is outside the circle.
(10) What is RF? Particle swarm optimization algorithm calculation process, parameter adjustment process?
(11) shortest path solution in dynamic programming;
(12) Optimization principle and objective of XGBoost;
13. How to measure the importance of a word;
(14) Query and matching of sensitive words;
(15) Transformer framework introduction;
(16) BERT’s main tasks, limitations; Why do I multiply by the scaling factor;
(17) The difference between the attention mechanism in Transformer and BahdanauAttention;
(18) How to handle negative sampling in Word2vec;
Feature engineering:
(1) What does feature engineering include?
(2) The treatment method of missing value;
(3) Detection method of outliers;
(4) The screening process of characteristic variables;
(5) Why normalization is needed;
(6) The difference between Spearman and Pearson;
(7) definition and treatment of over-fitting and under-fitting;
(8) How to solve the multicollinearity problem;
(9) Why do we need to do data sorting? The significance of processing continuous values into discrete values;
(10) Why to do model and business evaluation;
(11) What is the Chi-square test? Whether the IV value has been calculated;
(12) How to deal with the problem of gradient disappearance;
(13) The role of L1 regularization in logistic regression; Difference between L1 regularization and L2 regularization;
Linux:
(1) Check the contents and CPU instructions occupied by the model operation?
Tensorflow:
(1) 0 0 0 0 0 0
(2) How to speed up the process of online model;
04 Personal Reflections
When I was technically speaking, I kept a frank attitude. I would directly answer the interviewer’s questions, and honestly said no to the questions that I couldn’t. I also honestly said that I was playing with the two projects I had done in July.
After all, the interviewer sees a lot of people, ask a few more questions can test the candidate is not lying, and if you pass the interview by luck, in the future work if the boss according to your promise in the interview to ask you to complete more than your ability to do things, it is also very sad.
In addition, the attitude in the interview also reflects the attitude in the future post to a certain extent: “One: report the bugs that cannot be solved candidly to the team and the leader, and try to solve them together; Both: afraid of criticism, their cover finally cover caused greater losses; I think the former is better than the latter.
I will directly show my strengths and weaknesses in a square way. If the company can accept my weaknesses, I will make up for them in the future. If not, I will change to another company.
When I got the offer, I still had a lot of work to finish, such as brushing questions, resume modification, project optimization, course digestion and absorption, NLP knowledge summary and so on. Finding a job is not the end of the road, the algorithm is a long way to go, I hope we can become industry leaders and encourage each other.
Write at the end: Thank brothers let me and become attached to July, for July, let me be Dr Chu’s students thank Dr Chu took us like a master teacher, let me find direction in algorithm and personal pursuit, thanks to the employment of teachers and teaching assistants of July supervision and encouragement, thank the friends July answering questions, guidance and encouragement, meet you is the wealth of my life, thank you again!