Algorithm system

Question ReviewEven with a computational power system, computational power alone is not enough.

Computational power is merely the ability to compute; without the support of algorithms, computational power is zero.

It’s like this: you can’t just rely on gasoline to reach poetry and distant places; you must have a car, and then the gasoline can be put to use.

Lin Yuan, who was originally at a second-tier university level, was not proficient in algorithms.

Actually, this is easy to understand. If Lin Yuan had been inherently intelligent, he wouldn’t have scored a second-tier university in the college entrance examination back then.

Therefore, Lin Yuan’s ability to obtain an interview qualification for the computer major at Zijin Aviation School was entirely due to the guidance of the computational power system.

It was the computational power system that planned his learning path for him, found the right textbooks for each subject, and made Lin Yuan’s studies twice as effective with half the effort.

So, Lin Yuan had to face a fact: he was actually quite weak.

After all, IQ is not something that can be qualitatively improved by just reading a few books.

——

After two minutes, Lin Yuan’s antique T440 finally finished booting up.

If it had 630 security software installed at this time, it would definitely display in the bottom right corner: Your boot speed has surpassed X% of users nationwide (X must be a single digit).

As usual, Lin Yuan did not immediately open his Python development environment—PyCharm—after booting up. He gave the T440 another minute to ”cool down.” This was because an antique machine’s true boot completion often wasn’t as it displayed.

After a full three minutes, the PyCharm icon finally opened.

【loading project...】 Following this was a snail-paced loading process.

Subsequently, Lin Yuan finally opened his OCR recognition project.

That’s right, he directly used existing project code.

This was not him cheating.

In fact, just now, Professor Zhang specifically went to each classroom to say, ”This on-site assessment is different from before. Besides not allowing whispering or copying code, you are free to look up information online. You can refer to any open-source project.”

Professor Zhang even thoughtfully used magnets to stick a white paper with the school’s Wi-Fi name and password on the blackboard.

”But there’s one thing. Don’t be clever and think that if you modify someone else’s code, we won’t be able to tell. I’m currently working on a project that uses AI technology to determine code plagiarism, so I hope you don’t provide me with AI training material.”

Since even referencing open-source projects was not an issue, directly using one’s past projects was naturally even less of a problem.

However, what was meant by ”using” was far from simple copy-pasting.

In Lin Yuan’s view, all AI training processes could be abstracted as: data collection ==> data training ==> error calculation ==> error feedback.

Although this assessment was not explained in too much detail, Lin Yuan, relying on his diligent study over the past six months, had already completed the first step—understanding the problem.

The first step of any exam is always understanding the problem; many so-called difficult problems often have a large proportion dedicated to understanding the problem.

And judging by the furrowed brows of many people in this classroom, it seemed they hadn’t even passed the first hurdle.

The exam question asked everyone to implement OCR character recognition using original methods, but it did not specify in detail which categories counted as original. This was the first veil of this assessment.

If one were to truly be completely original from the ”data collection” stage, then they might as well give up and go to sleep.

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

But he had just looked around, and it was clear the school had no intention of keeping them overnight. The school didn’t even intend to keep them for dinner.

The time was roughly two in the afternoon, so the duration of this assessment would be three or four hours.

Lin Yuan thought to himself, ’This assessment is too casual; who knows which scoundrel came up with it.’

Since the assessment was only three or four hours long, it must have a key focus.

Lin Yuan pondered for a short while.

For OCR character recognition, collecting data features is a completely formulaic step. Because OCR recognition is not a new field, it certainly couldn’t be the key focus of the assessment. So, for this part of the code, copying it would be fine.

The remaining error calculation and error feedback also had nothing to innovate. More accurately, the school was holding a graduate entrance assessment, not recruiting national-level scientists. Were they still hoping to discover some genius through this assessment?

Therefore, the focus of this assessment must be on examining the candidates’ usual practice.

Simply put, it’s about examining what you usually do. Are you truly 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 very proficient.

And interest is the primary condition for scientific research.

Although Lin Yuan currently didn’t have much interest in scientific research, he was very interested in changing his fate by pursuing a career in scientific research.

Hmm, it could be considered different paths leading to the same destination.

Since the assessment is about daily practice and doesn’t expect you to be highly skilled, it’s naturally impossible to assess the error calculation and error feedback steps, because these two steps have too many ready-made routine solutions that don’t reflect individual differences.

That leaves only the last step—training data.

This step integrates linear algebra, calculus, computer programming ability, and image science, making it a very comprehensive step.

’It must be like this.’ Lin Yuan’s thoughts were set, and he immediately quickly located the ”training data” part of the code in the original project. He planned to refactor this part of the code, as for the other parts, they would remain untouched.

Lin Yuan then thought deeply.

Current AI training is nothing more than based on deep neural networks, but with only three-plus hours in total, it’s impossible to design a very deep network while writing code and debugging for results.

However, based on understanding the basic principles of images and mastering the basics of linear algebra and calculus, if one uses existing mathematical computing libraries, such as TensorFlow or PyTorch, to build a neural network of about three layers, three hours would be just about enough time.

At this point, Lin Yuan suddenly understood.

’Problem analysis complete. It turns out to be a comprehensive assessment question. The examiner is quite tricky; it essentially filters out those who just cram for exams. Graduate entrance exams at prestigious schools really aren’t just about doing problems.’

Thanks to his previous understanding of the OCR recognition project, Lin Yuan was confident he could complete the code writing and debugging in two hours.

The problem, however, was that this T440 couldn’t use a GPU for training.

The hardware entry requirements for modern AI training using mathematical tool libraries are too high; Lin Yuan had no choice but to switch to using the CPU for training.

He wasn’t worried about this, because the training process is continuous. As long as the error in the training results continuously decreases, even if time runs out, he could demonstrate the potential to convince the teachers to extend the time.

This is like climbing a mountain; even though you haven’t seen the cloud-shrouded peak yet, you can clearly tell from the feeling beneath your feet that your current location is not the summit, and there’s a higher place above.

Since the performance of each participant’s laptop in this assessment was different, Lin Yuan was certain: the focus of the assessment would definitely not be the length of the training time, but the accuracy of the final training results.

In other words: this assessment isn’t testing how fast you climb, but how high you can eventually climb.

Thinking of this, Lin Yuan couldn’t help but smile, ’No wonder Professor Zhang kept emphasizing the accuracy of the results earlier.’

His prediction perfectly matched Professor Zhang’s initial announcement of the evaluation criteria.

With his thoughts settled, Lin Yuan immediately began his busy copying + pasting.