Xueba's computing system

Chapter 4 Question Review
Even if you have a computing power system, computing power alone is not enough.

Computing power is just the ability to calculate. Without the support of algorithms, computing power is equal to zero.

It’s like: you can’t go to poetry and distant places just by relying on gasoline, you must have a car, and then the gasoline can work.

However, Lin Yuan, who is only at the second level, is not proficient in algorithms.

This is actually easy to understand. If Lin Yuan was born smart, he wouldn't have been ranked second in the college entrance examination.

Therefore, Lin Yuan was able to obtain the interview qualification for the computer major of Zijin Aviation School entirely due to the guidance of the computing system.

It was the computing system that planned the learning path for him and found the right textbooks for each subject, which made Lin Yuan's learning more efficient.

Therefore, Lin Yuan had to face the fact that he was actually quite incompetent.

After all, IQ is something that cannot be improved qualitatively just by reading a few books.
-
After two minutes, Lin Yuan's antique T440 finally started up.

If 630 security software is installed on it at this time, it will definitely be displayed in the lower right corner: Your boot speed beats X% of users nationwide (X must be a single digit).

As usual, Lin Yuan did not open his Python development environment - PyCharm immediately after the boot was completed. He gave the T440 another minute to "calm down". Because the actual boot completion of an antique computer is often not as it appears.

After three full minutes, the pycharm icon was finally opened.

【Loading project... 】What follows is a slow loading process.

Afterwards, Lin Yuan finally opened his OCR recognition project.

That's right, he directly applied the existing project code.

It's not that he cheated.

In fact, just now, Professor Zhang went to every classroom to say, "This computer test is different from the past. In addition to not allowing you to whisper and copy code, you are free to search for information online. You can refer to any open source project."

Professor Zhang was even thoughtful enough to stick a piece of white paper with the school's WiFi name and password on the blackboard with a magnet.

"But there is one thing. Don't be too smart and think that we can't tell if we just modify other people's code. I'm currently working on a project to use AI technology to determine code plagiarism. I hope you won't provide me with AI training materials."

Since it is not a problem to refer to open source projects, it is naturally even easier to directly apply personal past projects.

However, the so-called application is far from a simple copy and paste.

In Lin Yuan’s view, all AI training processes can be abstracted as: collecting data ==》 training data ==》 calculating error ==》 feedback error.

Although the test was not explained in detail, Lin Yuan had already completed the first step - reviewing the questions - by virtue of his hard work over the past six months.

The first step in any exam is always to review the questions. Many so-called difficult questions often require a large proportion of the exam time.

Judging from the frowning brows of many people in the classroom, it seemed that they had not even passed the first level.

The exam question asked everyone to use original methods to achieve OCR character recognition, but it did not specify in detail what categories were considered original. This was the first veil of this assessment.

If it is truly completely original from the "data collection" stage, then you can forget about it.

Lin Yuan roughly estimated that if every line of code was original from scratch, everyone would have to stay in this classroom overnight.

But he had just looked around and it was clear that the school had no intention of keeping them overnight. They didn't even plan to keep them for dinner.

The time now is roughly two o'clock in the afternoon, so the assessment will last about three or four hours.

Lin Yuan thought: "This assessment is too casual, who knows which idiot came up with it." "Since the assessment is only three or four hours, it must have a focus.

Lin Yuan thought carefully for a while.

For OCR character recognition, collecting data features is a completely formulaic step. Since OCR recognition is not a new field, it cannot be the focus of the assessment. Therefore, just copy this part of the code and you're done.

There is nothing innovative about the remaining calculation errors and feedback errors. To be more precise, the school is conducting a postgraduate entrance examination, not recruiting national treasure-level scientists. Do they expect to discover some genius through this examination?
Therefore, the focus of this assessment must be on examining the candidates' daily practice.

In simple terms, it is to examine what you usually do. Are you interested in computers and AI? If you are interested, then you will definitely write code frequently. If you write code frequently, you will naturally become proficient.

Interest is the first condition for scientific research.

Although Lin Yuan did not have much interest in scientific research at this time, he was very interested in changing his destiny through scientific research.

Hmm~~, it can be said that we have reached the same destination through different paths.

Since we are assessing your daily practice and not expecting you to have a high level, it is naturally impossible to assess the two steps of calculation error and feedback error, because there are too many ready-made routines for these two steps, and they cannot reflect the differences between people.

That leaves only one last step - training data.

This step combines linear algebra, calculus, computer programming skills, and graphics, making it a very comprehensive step.

"It must be like this." Lin Yuan thought about it and quickly located the code of the "training data" part of the original project. He planned to refactor this part of the code, and as for the other parts, he would leave them intact.

Lin Yuan continued to think deeply.

Today's AI training is nothing more than based on deep neural networks, but it only takes a little over three hours in total. It is impossible to design a deep network because you have to write code and debug the results.

On the basis of understanding the basic principles of graphics, mastering linear algebra and calculus, if you use the existing mathematical computing tool library, such as tensorflow or pytorch, to build a three-layer neural network, three hours is enough.

At this point, Lin Yuan suddenly understood.

'I have finished reviewing the question. It turns out to be a comprehensive test question. The person who wrote the question is quite bad. He has completely screened out all the candidates. It turns out that the entrance examination for graduate students in prestigious universities is not just about answering questions.'

Thanks to his past knowledge of OCR recognition projects, Lin Yuan was confident that he could complete code writing and debugging in two hours.

But the problem is that this T440 cannot use GPU training.

The hardware requirements for the modern mathematical tool library used in AI training are too high, so Lin Yuan had no choice but to switch to using the CPU for training.

He was not worried because the training process was continuous. As long as the error in the training results was getting smaller and smaller, even if the time was too short, he could get the teachers to agree to extend the time by showing his potential.

This is like when you are climbing a mountain. Although you have not yet seen the top of the mountain shrouded in clouds and mist, you can clearly know through the feeling under your feet that the place you are at the moment is not the top of the mountain, there is a higher place above.

Since the performance of each participant's laptop is different, Lin Yuan is certain that the focus of the assessment will definitely not be the length of training time, but the accuracy of the final training results.

In other words: this assessment is not about how fast you climb, but how high you can climb in the end.

Thinking of this, Lin Yuan couldn't help but smile, 'No wonder Professor Zhang just emphasized the accuracy of the results.'

His speculation perfectly matched the judgment results initially announced by Professor Zhang.

After making up his mind, Lin Yuan immediately began to busy himself with copying and pasting.

(End of this chapter)