I'm a scientist, why can't I have faith?

Is there a problem with the core architecture?

Upon seeing the comment, Wei Linfeng was filled with doubt.

"I remember we did a complete controlled experiment, didn't we?"

Wei Linfeng retrieved the data from the previous experiment.

The results show that global weighting is 2.1 percentage points more accurate than local constraint. The dual-path parallelism is also to limit the fault tolerance threshold, and regularization is to control overfitting. All of these have been fully verified.

In this situation, how can anyone still say that the core architecture of the algorithm is flawed?

Wei Linfeng immediately wanted to know who left the message.

"Who wrote this comment...?"

Upon closer inspection, there was no signature, but above this message was Lan Ce's message.

"Old Blue? Did he conduct another new experiment?"

If it were anyone else, Wei Linfeng would definitely have given them a good talking-to, but he didn't expect the person who left the message to be Lan Ce.

This is a big shot who has published a paper as the first author at CVPR. His level is no less than that of me, an ACM gold medalist, and may even be higher.

The former is a top conference in the field of computer vision, while the latter is the International Collegiate Programming Contest; both are highly prestigious achievements.

Of course, the two are very familiar with each other; they met when they were competing in the OI (International Olympiad in Informatics) in high school.

"I remember Lao Lan should be preparing his Caltech application... how could he have time to optimize this competition algorithm?"

Su Yuran's team has an impressive lineup, including an ACM gold medalist and a first-author CVPR author. Even for a preliminary round question, they have academicians to endorse it.

It's no exaggeration to say it's like using a cannon to kill a mosquito.

The only reason she invited Wang Dong to join was to meet the requirement of participating in the inter-college competition.

However, it was this non-professional who was brought in to make up the numbers who answered the questions about Wei Linfeng's graduation project, a gold medalist in ACM.

Since it was most likely a message written by Lan Ce, Wei Linfeng subconsciously reconsidered the algorithm design once again.

"If we follow Lao Lan's idea... then regularization should be placed after feature extraction, not in the inference branch..."

"Create a dynamic trigger for dual-path redundancy, so that it only starts under extreme operating conditions..."

As an ACM gold medalist, Wei Linfeng quickly developed a new algorithm based on the optimization suggestions in the comments.

From seeing the comments to the birth of the new algorithm, only about a hundred lines of code were modified, and the whole process took no more than twenty minutes.

After selecting the standard test set provided by the "East Asia Smart Cup" organizers and their own supplementary extreme working condition test set, Wei Linfeng stretched and waited for the test results of the new algorithm.

Three sets of numbers appeared on the screen shortly afterward:

Congestion prediction accuracy: 99.62%

End-to-end inference latency: 37ms

System fault tolerance: 0.08%

"Holy crap, Old Blue is awesome!"

After optimization, the algorithm improved accuracy by nearly 1 percentage point, reduced latency by more than half, and lowered the fault tolerance to less than one-fifth of the required level.

I only changed about a hundred lines of code and adjusted the order of some programs, yet there was such a big improvement.

More importantly, Wei Linfeng's stagnant graduation project has a new direction.

Although not exactly the same, the optimization ideas in this message can be fully applied to his graduation project...

The fatigue accumulated during the National Day holiday vanished, and Wei Linfeng felt an unprecedented excitement as he seized upon this fleeting moment of inspiration.

He first saved the optimized algorithm for the preliminary round, and then turned around and started looking for information according to his new ideas.

By the time he came to his senses, it was already the next morning.

Wei Linfeng unknowingly spent the entire night in the laboratory.

However, he felt no fatigue at all; instead, he felt a sense of relief and joy.

His graduation project... is finally making progress.

Based on the inspiration drawn from the comments and his overnight research, Wei Linfeng's graduation project had taken shape.

The next step is to write the code according to the proposed approach and then conduct experimental testing.

If everything goes according to his plan, all that's left is tedious paperwork.

Even if I slack off a little, I can still finish this work within a month.

We're almost there!

Wei Linfeng was so happy that he completely forgot about joining the crew ahead of schedule.

Choosing an academic path means you can't be considered to have made it through until you receive your master's or even doctoral degree.

This was an extremely arduous and long journey.

However, it's a bit too premature to think about these things now. After all, conservatively speaking, it will be another seven or eight years before Wei Linfeng makes it big.

However, it is certainly worth celebrating that the problem of the graduation project has been solved.

Wei Linfeng's first thought was to thank the author of the comment. If it weren't for that comment, it might have taken him much longer to find a concrete solution.

Ignoring the fact that it was six o'clock in the morning on a weekday, Wei Linfeng took out his phone and made a call to Lan Ce.

The phone rang for quite a while before Lan Ce reluctantly picked up the phone that was charging on the bedside table.

He initially thought it was an alarm clock, but his fingers moved instinctively according to muscle memory, and he ended up answering a call and turning on speakerphone.

A loud and extremely excited voice came from the other end of the phone.

"Old Lan, you're a fucking genius!"

The sudden roar from the lioness from Hedong instantly woke Lan Ce up.

Holy crap, what's wrong with Wei Linfeng so early in the morning?

Is this person driven crazy by their graduation project?

Lan Ce knew that Wei Linfeng had been worrying about his graduation project these days, so he deliberately didn't contact him much.

Now look what's happened, this kid is bothering him first thing in the morning.

You should know that Lan Ce didn't have classes today. He was working on documents until 2 a.m. yesterday, and he was planning to sleep in today.

As a result, I had to get up even earlier than 8 a.m.!

Ignoring Lan Ce's bewildered state, Wei Linfeng continued talking to himself.

"I didn't expect you to care so much about my graduation project. The comments in that document were really inspiring..."

"But next time something like this happens, you can just send me a private message. No need to be so subtle. We've practically been through thick and thin together..."

Although his brain wasn't fully activated, Lan Ce still keenly sensed the key point.

Documentation? Comments? Algorithm optimization?

He's never done any of these things before.

Lan Ce never opened the "East Asia Smart Cup" app again after finishing the preliminary algorithm. How could he possibly leave a message?

"I think you've really learned to be a fool, Wei Linfeng!"

"I didn't do any of this, why are you calling to thank me?"

This time, Wei Linfeng on the other end of the phone was confused. It wasn't Lao Lan's message.

Who else has that level of skill?

It wasn't that he underestimated Su Yuran, but the fact that she could notice such a subtle point of optimization meant that her understanding of algorithms was definitely no less than his and Lao Lan's, and perhaps even a step above theirs.

Although Su Yuran is indeed capable, as a sophomore, she doesn't yet possess that level of understanding.

If she has this level of skill, she can definitely be called a genius; she wouldn't even need me and Lan Ce to join this competition.

Could it be that Academician Du Fan wrote it?

I don't think Academician Du Fan should be such a free person...

Not only Wei Linfeng, but even Lan Ce, who was inexplicably called out by him, was a little curious.

The first thing he did after hanging up the phone was to open the shared document.

At this point, the algorithm in the document had been updated by Wei Linfeng, and Lan Ce naturally saw the data after the algorithm was optimized.

This improvement is no small matter.

He finally scrolled to the unsigned message.

Not only did it pinpoint the problem accurately, it also provided specific solutions.

Simple, efficient, and convenient.

That's an absolutely brilliant comment.