The Expert Was Practically a Cheat Code

[I'll take a look and get back to you.]

Beibei: [Mm-hmm, no rush, expert. I'll go study your paper first!]

After closing the chat window, Wu Chen opened the paper Chen Beitao had written and studied it closely.

Besides theory, it contained a great many calculations and models, so he had to work through the calculations while writing code to verify them.

It was not until four in the morning that Wu Chen finally grasped the central idea of the entire paper.

It still fell a little short of a true residual network, but only a little.

Chen Beitao had not taken that final step, not because she lacked ability, but because of the limitations of the era.

After all, in 2008, the vanishing gradient problem in deep neural networks had yet to be solved, and computing resources were nowhere near sufficient to train ultra-deep models.

Even AlexNet had yet to appear, let alone residual networks.

A theory in which academia could see no future and a line of research that offered investors no returns were destined to remain no more than a passing fancy that had struck a genius.

On the original timeline, Chen Beitao would never have realized what she had accomplished.

If Wu Chen had not brought it up, she probably would not even have remembered researching it.

But now, Wu Chen had added the final missing dot to the dragon's eye.

The identity mapping algorithm.

It could create shortcut connections, allowing gradients to pass through the heart of an ultra-deep network without loss. It could be said that without it, no network beyond a certain depth could be trained effectively, and the depth of large AI models would forever be confined to shallow layers.

After doing all that, Wu Chen called up the system panel.

[Host: Wu Chen]

[Current host's basic information:]

[Intelligence: Level 1 (15% -> 25%)]

[Stamina: Level 1 (1%)]

[Skill levels]

[Artificial intelligence: Level 1 (0%) -> Level 1 (5%)]

[Materials science: Level 0 (0%)]

[Chemistry: Level 0 (0%)]

[Physics: Level 0 (0%)]

[Engineering: Level 0 (22%)]

[Energy science: Level 0 (0%)]

[Sample Shop: Vengeance No. 1, Little Jiang No. 1, Dawn No. 1, Little Jiang No. 2]

[Available production blueprints: Dawn No. 1, Little Jiang No. 1, Little Jiang No. 2]

[Skill Points available: 2.827]

As expected of a genius expert.

If he had researched and developed a residual network from scratch himself, it would probably have taken him more than a week.

Now, with something ready-made to work from, he had increased his intelligence by 10% and his artificial intelligence skill by 5% in just a few hours.

She was practically a cheat-code accelerator for him.

After some thought, Wu Chen sent the completed document to Chen Beitao, then sent over a list of several theories he had not studied in his previous life.

[The theoretical convergence of variational inference in generative models, optimization algorithms from graph theory for biomolecular network modeling, the implications of random matrix theory for neural network initialization, and applications of information geometry in analyzing model expressiveness. Look into these directions when you have time.]

After sending that, Wu Chen thought for a moment and added another message.

[Don't spend time on SNN anymore.]

Chen Beitao did not reply. He did not know whether she was asleep or absorbed in her research.

Wu Chen paid it no further mind and instead pondered his 2.827 Skill Points.

Little Jiang No. 2 had to be put into production quickly, so he could monetize the technology as soon as possible and obtain more Skill Points.

When the time came, buying samples or production lines with them, or even adding the points directly to his personal attributes, would all be good options.

If he piled on Skill Points like crazy and raised his stamina to level 10, could he punch straight through Wang Hongfa's dog head?

Putting that strange thought aside, Wu Chen washed up and went to bed.

Over the next few days, he did his best to maintain a regular schedule.

He got up at six every morning to run and went to bed at ten at night, getting his disrupted body clock back in order.

After all, a semester of eight o'clock classes was no joke.

If he was late to class after the semester began, it would be a major teaching incident.

If Wang Hongfa caught him two or three times, Wu Chen could basically pack his bags.

Still, such a regular life had its benefits.

His stamina level rose slowly, at a rate of 1% each day.

But his biggest surprise was Chen Beitao.

This true expert had actually given up SNN and turned to researching the several directions Wu Chen had given him.

Then, in less than a week, she managed to complete research on predicting molecular properties based on graph convolutional networks and proving the convergence of variational inference.

When she sent Wu Chen the two papers, there was a hint of shame between the lines.

Beibei: [Expert, expert, I'm sorry. I'm really too stupid. I only got this little bit done in a week. I'll definitely work harder next week!]

[...Keep it up.]

For a moment, Wu Chen did not know what to say. He could only quietly tell her to keep it up.

If he had done it himself from scratch, completing that much in three months would already have been lightning fast.

After absorbing her research, Wu Chen's artificial intelligence level rose again.

[Artificial intelligence: Level 1 (5%) -> Level 1 (10%)]

If Chen Beitao could produce such high-quality work every week, wouldn't he be able to level up while lying down?

Unfortunately, Chen Beitao had a job of her own, and she had only been so quick because these two directions happened to be ones she had researched before.

The other research would have to be worked through slowly.

In any case, Chen Beitao's research into these two key algorithms, combined with Wu Chen's own research into positional encoding, had finally completed another crucial step on the path from word2Vec to the Transformer architecture:

The mathematical foundation of the attention mechanism!

Completing it was equivalent to clearing the theoretical obstacles to putting the attention mechanism into practice, giving core components such as multi-head attention and positional encoding rigorous mathematical support rather than leaving them as mere engineering experiments.

Once the major theories of parallel computation optimization for multi-head attention, efficient implementation of positional encoding, engineering improvements to layer normalization, activation function adaptation for feedforward networks, and cross-layer attention interaction mechanisms were completed, the artificial intelligence of 2025 could appear ahead of its time!

When that day came, he would personally announce the arrival of the age of artificial intelligence to the world!

Of course, this was no easy task.

But if he could gather a group of smart people like Chen Beitao, that future should not be too far away.

Before Wu Chen could embark on his great voyage, however, he had no chance of meeting that many smart people. Instead, he had to get along with a group of adorable first-year students.

Whether he was running a company or submitting papers outside school, he still had to face his core duty as a teacher:

Teaching and nurturing students.

In fact, students who could get into Xiaoheshan University of Technology had no trouble with their ability to learn.

But for the teachers in the computer science department, this was a big problem.

Many students did not even know how to use a basic computer!

This happened in two periods.

One was the early years of the twenty-first century, when people were poor. Apart from going to internet cafes, not many families could afford to buy a computer for their homes, let alone students from remote areas.

The other was twenty years later, when quite a few university students knew only how to play on their phones and could not use computers.

The C programming course Wu Chen taught faced the first situation.

In five periods that morning, out of sixty students across two classes, only three had set up a basic development environment.

Wu Chen was not impatient, but when class ended, he felt a trace of helplessness and wanted to sleep.

Being a responsible teacher was not easy, so some people chose either to slack off or to focus on research.

Teach classes?

Couldn't they just leave that to the teaching assistants?

If they had to choose, publishing papers and securing research funding were the top priorities.

Wu Chen would never do such a thing.

At noon, he grabbed a quick bite at the school cafeteria, then went early to the lab where he would be teaching.

The afternoon students were in their third year, so things should be much better.

After arriving, Wu Chen switched on the teaching computer at the lectern and began preparing.

He had only been at it for a short while when footsteps sounded outside. A student had arrived for class.

Wu Chen glanced at the time in the lower right corner of the computer screen.

12:50.

The afternoon class started at 1:30. Third-year students were already old hands and rarely came to the classroom this early.

It looked like he had found a good student.

His thoughts wandering, Wu Chen looked up at the door.

And found himself looking into a pair of bright eyes.

What a beautiful girl!

Once he saw her face clearly, even Wu Chen could not help admiring her inwardly.