Deep Space Wandering Fleet

Chapter 74: Chapter 74: You Have Met a True Pervert

“Please take 3 minutes to think,” said the moderator.

Soon, 3 minutes later, the host handed the microphone to Cai Kun.

“Next, please let Classmate Cai answer.”

His face turned red as he stumbled over his words for a long time, “I’m sorry, I have not studied this field in depth… A recursive neural network is an artificial neural network that has a tree-like hierarchical structure and recursively processes input information according to the sequence of its connections. It is one of the deep learning algorithms.”

Having reluctantly recited a few definitions and struggling to convince the audience, Classmate Cai said, “I’m sorry, I don’t have any particularly good ideas at the moment.”

A round of jeers came from below, but Classmate Cai, despite his flushed face, still bravely stood on the stage.

He looked up at Zhang Yuan.

Let’s see what you can answer. If you can’t answer anything, how can you score higher than me in the interview?

“Next, please let Classmate Zhang answer.”

“I have a question,” Zhang Yuan raised his hand. “I’ve heard about Hopfield networks, but what is network stability?”

The questioner smiled and explained, “If certain weights within the network can converge to an equilibrium point, which is called weight convergence, then the output can also approximate the desired output, which is called system stability. Hence, convergence pertains to variables, while stability relates to the system.”

“The system’s stability is not guaranteed, the control system is unstable, and the network’s convergence loses its foundation.”

Zhang Yuan thought for a moment and said, “I have not studied your research field in depth either, only read some papers. But I have some basic ideas you might be interested in hearing.”

“…Firstly, the issue of monotonicity, the Hopfield network model in discrete time with continuous states depends on whether neurons’ activation functions are monotonic, or at least locally monotonic.”

“Secondly, by studying the condition under which the energy function becomes a convex function, the operation of the Hopfield network can be viewed as a constrained convex optimization problem solution, thereby attempting to prove the sufficient condition for a globally unique minimum point…”

“Convex optimization, you are correct! Convex optimization is exactly the direction I’m working on. I have another question…”

This time, the question resonated deeply with the questioner, who then asked a few more questions related to convex optimization, sparking a mathematical discussion on the spot.

Finally, at the host’s cue, Zhang Yuan reluctantly said, “Regarding knowledge about convex optimization, if needed, you can discuss with me backstage later; I’d prefer not to go into detail here…”

Much of what he said was just blabbering based on his usual accumulation of knowledge; if he could spontaneously write a paper on the spot, he wouldn’t need to be on stage; he might as well be hailed as a “paper God.”

A burst of applause rose from below.

In such a short time, the outcome was clear.

Cai Kun wished he could vanish into thin air.

All he could do was redden his face, quietly encouraging himself, perhaps due to the limited sample leading to variance—he just happened to face a topic he was unprepared for but his opponent knew.

The host then asked, “Contestant number 666, please state your question.”

The questioner was a girl and her question was more practical, not so obscure, “I want to ask about caching in servers.”

“Caching is divided into two phases: the first phase is the data placement phase, where communication resources are used in non-peak demand times to place data in each user’s cache. The second phase is data delivery, assuming that during peak demand times, each user randomly requests a complete file, and the server justly distributes complete data to meet all users’ needs.”

“My question is, how can we scientifically design the caching scheme?”

The questioner even projected a PPT onto the screen.

It was a specialized question, but everyone understood what was being asked.

The host said, “Alright, please think about it for 3 minutes.”

The microphone was handed to Zhang Yuan.

Smiling, he said, “I think… let’s give the chance to Classmate Cai first. I’m afraid if I speak now, he’ll have nothing left to say.”

Laughter erupted from below, and even some whistling sounds.

Some believed him, others did not.

Cai Kun was startled; this time, he did have some ideas, whereas Zhang Yuan seemed at a loss, needing more time to think.

“…I have this idea, each user caches a portion M/N of each file, and during the data delivery phase, the server releases the missing (1-M/N) portion for each user individually, making the transmission rate R=K(1-M/N)…”

“Has anyone heard of a scheme called ‘coded caching’? The specific algorithm is like this…”

“You make a good point.”

Although the questioner courteously responded, there was a slight hint of disappointment. Coded caching was a quite mundane idea with established algorithms already, bearing no research significance. Bringing such a worn-out idea for a project would likely get one slapped by their supervisor.

It was Zhang Yuan’s turn next and he cleared his throat.

“Classmate Cai’s idea is good. In the distribution phase, by utilizing the relationships among existing cache information, it is feasible to design certain coded combinations that enable multiple users to decode the needed information from a single broadcast, thus achieving global caching gains.”

“However, a primary limitation is the need to split each file into segments exponentially growing with the number of users, K. As we know, slicing at an exponential scale is daunting to execute in algorithms…”

“May I use your digital pen and the big screen?”

The host paused, “Sure.”

Zhang Yuan drew several diagrams on the screen.

“…Essentially, this is a combinatorial graph theory problem. What we want to achieve is a caching scheme where F is a polynomial level of K and R is at a constant level or prove whether such a scheme exists or not.”

“Obviously, whether file segmentation can be converted into polynomial levels, i.e., to an industrially usable scale, relies entirely on whether one can construct a hypergraph problem that satisfies certain properties, which are as follows…”

“I think after converting it into a mathematical problem, it might make a decent paper. As for the solution, I certainly cannot provide it now; it’s just an idea for now. How to tackle this graph theory problem is something this classmate ought to ponder on his own.”

Whispers filled the room.

Among the audience, many of whom were professionals, whether what was said made sense was naturally discernible to them.

Applause broke out!

A few minutes later, the questioner led the applause.

Thunderous applause cascaded down.

Cai Kun could no longer face staying any longer; his face entirely flushed, and he quickly nodded in concession.

He finally realized that although both scored 100 points on an exam, his were from ability, whereas others scored the same because the paper only awarded up to 100 points.

“Old Cai, you encountered a real freak!”

“Yeah…”

“It seems like a fair loss.”