[PYTHON] 2nd AI Implementation Test Class A Pass Experience
I have successfully passed the 2nd AI Implementation Test Class A (implemented on September 26, 2020), so I will leave a record of my experience! It's my own experience, so don't be afraid.
The score is Mathematics: 95% Python : 100% AI : 100% was.
Pass notification email
What is AI implementation test class A?
From Official Site
I have a basic knowledge of mathematics and programming about the implementation of deep learning, I can start reading theoretical books on deep learning, and I am ready for self-study. In addition, it is a level where you can challenge the certification program of E qualification (sponsored by Japan Deep Learning Association), which is currently the highest peak of AI qualification examination.
- Even if you pass the AI implementation test A grade, you do not have the right to receive the E qualification. You must take the E qualification program.
About the author
--Third year engineer ――Major in electronic engineering at university, graduated from graduate school --As a related qualification, you have obtained the G certification / Python3 engineer certification basic exam.
difficulty
Information on the qualification examination information site The above site considers the following (information as of October 2020).
AI implementation test A grade is a subordinate qualification of E qualification, and the difficulty level is lower than G test. Compared to other AI-related qualifications, it is more difficult than the Python3 Engineer Certification Basic Exam and easier than the Python3 Engineer Certification Data Analysis Exam.
As for the impression that the G test and the Python3 engineer certification basic test have been obtained, I think it is a fact (although the question area is different) that "the difficulty level is lower than the G test". The AI implementation test A grade is much easier to study than the G test in that it is narrowed down. I generally agree that it is more difficult than the Python 3 Engineer Certification Basic Exam. For those who have a natural knowledge of mathematics, the AI implementation test A grade may seem easier.
In addition, the number of successful applicants and the average score rate this time are as follows. (Official Information)
Number of examinees: 181 Number of successful applicants: 133 (Approximately 73% when calculating the pass rate)
Average score rate (overall) 74.63% Average score rate (math) 71.33% Average score rate (python) 75.14% Average score rate (AI) 77.43%
The pass rate is lower than the first time (209/258, which is about 81% when the pass rate is calculated).
The pass line has not been announced, It is said that 80% is required to pass the test without fail.
Second question range
The range of questions at the time of the second session was as follows. (For some reason, the information on the official website is copied here)
20 AI titles
--Input layer and output layer
- Weight --Calculation of forward propagation --Matrix multiplication --Introduction of bias term --sigmoid function --Introduction of correct answer value --Square sum error --Differentiation of error --Error back propagation method --Chain rule --Partial differential --Hadamard product
20 programming subjects
- Numpy
- Pandas
- Matplotlib
- Seaborn
- Sciket-learn
20 math subjects
--Sets and probabilities-Union and intersections-Absolute and relative complements-Bayesian probabilities-Conditional probabilities --Sequences and matrices -Question the reading comprehension of mathematical formulas required to describe the basic network of neural networks --Functions and derivatives --Question the reading comprehension of mathematical formulas used in the chain rule of neural networks
Examination system
As you may know, it is ** online exam **. Take the home exam. You can take the test while checking. However, since 60 questions are asked per hour, it is a delicate point whether there is time to investigate. As in the official example, the questioning method is an alternative question of choosing one from four options.
Below is a bulleted list of what I thought about the examination system.
――When you select an option, you will hear a "picone" sound for some reason. --There is no ** check for uneasy questions and features that you can look back on later, which are common in other exams. (Press the upper left to go to the problem list) --The problem statement cannot be copied.
Study method
(I'm just writing what I did. I don't think it's the best way to study, so please use it as a reference.)
I mainly did the following udemy teaching materials. Deep Learning: Artificial Intelligence (AI) and Deep Learning Principles Built and Learned from Zero with Python
I listened to the explanation while copying the contents of the lecture on the Jupyter Notebook. I did it steadily from Section 1 to 7, and from Section 8 I did not do it because it was not in the [2nd question range](# 2nd question range) in the convolutional neural network. It is very easy to understand with teaching materials that can implement deep learning from 1. Recommended regardless of the exam.
Also, since the above teaching materials had only the minimum necessary math content, the same instructor Mathematics course for AI: Linear algebra / probability / statistics / differentiation for artificial intelligence learned little by little I also bought.
I strongly recommend you to buy it because it covers the first teaching materials and contents. Section7 Probability / Statistics Last "Conditional Probability and Bayes' Theorem" was insanely useful in Implementation Test Class A.
Please note that this study method should be used as a reference, but ** the above materials alone do not cover the range of questions ** Question range (official site) It is necessary to make up for the missing part by staring at it.
What was not enough in the second question range
- Pandas
- Seaborn
- Sciket-learn etc...
About preparation and solution on the day
You should prepare your writing utensils and calculation paper. I have the impression that it was calculated numerically more than I imagined.
You can do anything on your PC, but you may also want a calculator.
(I think it would be useful to start Python if anything)
The problem statement is relatively long, but I didn't read much in the first half because I didn't have much time. There are many simple calculation problems, so you can solve them without reading.
Impressions after taking the exam
I solved it in order from the front, and it took only about 5 minutes. If you know the matrix calculation and partial differential, you can solve about half.
1st information and this time
Reference article about the first AI implementation test class A Since I passed the 1st AI implementation test [A grade], I tried various things Impressions and memorandum of AI implementation test
Especially about what is written in the second article, I will describe the current situation at the time of the second article.
In the notes of the test, "In the test, pressing the" Start "button after [14:05] and pressing the" End "button after [15:00] will not record the execution history in the system and will not be executed. Please note that it will be "failed". There was a description. I understand that admission is only allowed 5 minutes after the start of the test, but I wonder what it is like to be able to press the "Start" and "End" buttons even after the test ends.
The system doesn't seem to be updated as the same announcement was made the second time.
I'm also worried that the achievement rate was 0% after answering all the questions. I hope it doesn't mean that everyone failed on the day of the announcement of the results.
I can't log in to the test site anymore, so it's not accurate information (sweat) I remember an announcement on the site saying "Achievement rate is dummy" near the exam. I didn't remember what happened because I thought "that's right" and ignored it during the exam. (Probably this is also the same specification)
** Comprehensiveness of official text ** I haven't bought it this time, but I don't think there is any deviation from the table of contents of the teaching materials (naturally). However, even if you compare the table of contents of the teaching materials with the question range, you can see that "set and probability-union and intersection-absolute complement and relative complement-Bayesian probability-conditional probability" seems to be missing. It seems that you need to study separately here.
Digression
It's a selfish image, but I feel that online-style exams are evolving quickly. (I hear that the question format of the same online G test has changed considerably compared to the past) I wonder if AI implementation test A grade will evolve further from now on. (On the contrary, it would be a problem if the system side does not evolve)
I thought that the question range would change with each round, so I just copied and described the [second question range](#second question range) this time.
In that sense, the [study method](# study method) in this article is likely to be tentative. There are some places where the price has changed due to campaigns etc., but I think the official textbook is the safest way to study.
Finally
I wrote a lot, but I think that if you firmly establish the basics of deep learning, you will definitely be able to take it. (Of course, there are individual differences in understanding mathematics.)
I myself was worried because I didn't have much information about my qualifications, but now I can tell myself in the past that I don't have to be afraid more than necessary (laughs).
** Last Last ** There is a page called "AI Implementation Test Goal" on the official website, but I would like to introduce it because there was such a passage there.
I would like people in the humanities rather than sciences to take on the challenge, and those who have less environment and motivation than those who have no inconvenience in the study environment.
I think it's emo without permission (laughs)
Rather than ending AI with "vaguely amazing things," it may be possible to come up with various ideas and improve the world by increasing the number of people who can say and understand how AI works.
I think that those who are reading this article are mainly those who are going to take the AI implementation test A grade, so please do your best to challenge everyone! I hope you find this article helpful, even if only for a moment.