Deep Space Fleet

"Two contestants are asked to think for 3 minutes."

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

"Next, please ask Cai to answer."

He blushed and stumbled for a long time: "I'm sorry, I haven't studied this field in depth... A recurrent neural network is an artificial neural network with a tree-like hierarchical structure and network nodes that recurse input information according to their connection order. One of the deep learning algorithms."

Reluctantly recited some definitions, but was really confused, Cai said: "I'm sorry, I don't have a particularly good idea at the moment."

There was a burst of booing from below. Although Cai blushed, he still stood on the stage bravely.

He looked up at Zhang Yuan.

Depends on what you can answer. If you can't answer anything, why do you have a higher interview score than me?

"Let's ask classmate Zhang to answer."

"I have a question." Zhang Yuan raised his hand, "I've heard of the Hopfield network, but what is network stability?"

The questioner smiled: "If some weights of the network can converge to an equilibrium point, that is, the weights converge, then the output can also fit the desired output, that is, the system is stable, so the convergence is for the variable Stability is for the system."

"The stability of the system cannot be guaranteed, the control system is unstable, and the convergence of the network has lost its foundation."

Zhang Yuan pondered for a while, and said, "I haven't studied your research field in depth, I just read some papers. But I have some simple ideas, you can listen to them."

"...The first is the problem of monotonicity, whether the activation function of the neuron in the discrete-time continuous state hopfield network model is a monotonic function, or whether it is locally monotonic."

"Secondly, by studying the conditions for the energy function to become a convex function, the operation of the Hopfield network is regarded as the solution of a constrained convex optimization problem, so as to try to prove whether there is a sufficient condition for a globally unique minimum point..."

"Convex optimization, you are right! Convex optimization is exactly what I am doing, and I have another question..."

This time, it touched the heart of the questioner, and he asked a few more questions about convex optimization, and exchanged mathematics on the spot.

Finally, under the hint of the host, Zhang Yuan had no choice but to say: "If you need knowledge about convex optimization, you can discuss it with me behind the stage. I won't go into details here..."

He is just talking nonsense about many things based on his usual accumulation. If he can really write a paper according to the circumstances, then he doesn't need to be on stage, and he can just be the "thesis god".

There was already a burst of applause from below.

In such a short period of time, the judgment is judged.

Cai Kun wished he could find a crack in the ground and get in.

He could only blush and cheer himself up secretly, perhaps because of the difference caused by too few samples, he happened to encounter a problem that he could not but the other party knew.

The host asked again: "No. 666, please tell me your question."

The questioner is a girl, and her question is relatively pragmatic and not so off-the-wall: "I want to ask a question about caching in the server."

"Caching is divided into two stages: one is the data placement stage. During the idle period when the data demand is small, use the communication resources with spare capacity to place data in each user's cache device. The second is the data distribution stage. Assume that during the peak period of data demand, each user randomly requests a complete file from the server, and the server comprehensively considers these needs and distributes complete data to meet the needs of all users."

"My question is, how to design a caching scheme most scientifically?"

The questioner even sent a PowerPoint to the screen.

This question is very technical, but everyone can understand what is being asked.

The host said: "Okay, everyone, please think for 3 minutes."

The microphone was handed over to Zhang Yuan.

He said with a smile: "I think... let's give the opportunity to Cai Kun first. I'm afraid that after I say it, he won't have anything to say."

The people below laughed again, and whistled.

Some people believed it, and some people didn't.

Cai Kun was taken aback for a moment. This time, he still had some ideas, but Zhang Yuan who was opposite seemed to be unable to say anything and needed more time to think.

"...I have such an idea that each user caches the M/N ratio data of each file separately, and in the data distribution stage, the server distributes the missing (1-M/N) data of each user one by one. At this time, the value of the transmitted data R=K(1-M/N)..."

"I don't know if you have heard of a scheme called 'coding cache'. The specific algorithm is like this..."

"You have a point."

Although he said so, the questioner was slightly disappointed. Encoding caching can be said to be the most common idea. There are already mature algorithms, and there is no research significance. If you use this bad idea to start a project, you will probably be slapped back by your instructor.

Next it was Zhang Yuan's turn, he cleared his throat.

"Student Cai Kun's idea is very good. In the distribution stage, use the relationship between the existing cached information to design a certain encoding combination of the required broadcast content, so that multiple users can simultaneously decode from a single information. part of the information needed, resulting in a global cache gain."

"But its limitation is that each file must be divided into a parameter that grows exponentially with the number of users K. As we all know, exponential division is difficult to achieve algorithmically..."

"Can I borrow your digital pen and big screen?"

The host froze for a moment, "Yes."

Zhang Yuan drew a few pictures on the screen.

"...Essentially, it is a problem of combinatorial graph theory. What we hope to get is a caching scheme when F is a polynomial level of K and R is a constant level. Or prove the existence of such a scheme."

"Obviously, whether the number of file divisions can be converted to a polynomial level, that is, to an industrially usable level, depends entirely on whether a hypergraph that satisfies certain properties can be constructed mathematically. These properties are like this..."

"I think that after converting it into a mathematical problem, I can already write a good paper. As for what the answer is, I definitely can't answer it now, it's just an idea. How to solve this graph theory problem depends on this student. Think for yourself."

There was a murmur from below.

There are still many professionals here who eat melons, and they naturally have their own judgments about whether they are talking nonsense.

Clap clap!

After a few minutes, the questioner led the applause.

There was thunderous applause from below.

Cai Kun no longer had the face to stay any longer, his face was flushed, and he nodded hastily to admit defeat.

He finally knew that the same 100-point test paper, he got 100 points because of his strength, while others also got 100 points because there were only 100 points on the paper.

"Old Cai, you have met a real pervert!"

"yes……"

"It seems that the loss is not wronged."