Chapter 5: Excuse me, I must interrupt.

After four students had finished, it was soon Zhou Yun’s turn.

“Hello, Zhou Yun. You haven’t brought a resume. Why don’t you start with a self-introduction and then share your plans for the next three years?”

Zhou Yun nodded. “My name is Zhou Yun. I graduated from...” He briefly introduced his alma mater and his preliminary exam results. The self-introduction was over quickly; after all, the past twenty uneventful years offered little worth mentioning.

“As for the next three years, my plan is to focus seriously on research and strive to pursue a doctoral degree...”

When Deng Yonghua heard the words “doctoral degree,” his eyes lit up as if a keyword had been triggered.

Qiu Yan and Shen Rui, sitting beside him, looked at Zhou Yun with a hint of admiration.

This kid was quite bold—going for a doctorate? That’s not something just anyone can do.

After Zhou Yun finished speaking, Deng Yonghua nodded. “I just heard you say you’re very interested in research. Although you answered all five questions correctly, relying on that alone is far from enough to excel in research. What do you think are your strengths in this field?”

Answered all five questions?! The expressions of Qiu Yan, Shen Rui, and the four students sitting nearby changed noticeably.

Qiu Yan and Shen Rui thought: This junior fellow is impressive. If nothing unexpected happens, the professor will definitely take him on. Just his foundation alone is uncommon.

Meanwhile, the other four thought: Oh no, we’ve met a big shot! They couldn’t help but feel a surge of anxiety.

“Professor, I actually think I have some talent for research. But you might not believe me just by my words, so I prepared a PowerPoint presentation. May I briefly report on it?”

When speaking with a research-oriented mentor, honesty is the best policy.

Deng Yonghua raised an eyebrow. “Sure, of course. We have a wired connection here. Did you bring your laptop? If not, I can lend you mine temporarily.”

“I brought my laptop.” Zhou Yun took it out from his backpack and skillfully connected it to the large screen via HDMI.

As the laptop booted up, the four students by the wall felt as if they had been struck again.

Wait, this isn’t just an interview? Why is he giving a presentation now?

They didn’t know what Zhou Yun planned to present, but in this setting, no one would dare speak casually without something real to show.

Zhou Yun’s standing in their estimation rose another level.

Qiu Yan and Shen Rui’s expressions grew more complex; their glances toward Zhou Yun were tinged with sympathy.

Although Deng Yonghua was usually kind, when it came to academic matters, he was extremely strict. Being scolded by him was a standard experience.

At every group meeting, if no one got scolded, that would be truly strange.

Despite Deng’s current smile, soon he would make you understand what rigorous research truly means.

Zhou Yun rose and walked to the large screen beside him.

He was about to start his... no, his report.

“AgileEdge: Adaptive Co-Optimization for Pervasive Low-Latency Edge AI. This is the topic I will report on today.”

His fluent English earned Deng Yonghua’s satisfied nod, though his expression turned significantly more serious.

Before Zhou Yun could continue, Deng asked, “Are you reporting on a paper? Which journal or conference is it from? Who are the authors?”

Qiu Yan and Shen Rui wore “just as expected” looks. They recalled that during their first presentations, despite some beginner’s protection, Deng had grilled them mercilessly.

But when they looked at Zhou Yun, they noticed not a trace of nervousness on his face.

Indeed, after three years of countless group meetings, Zhou Yun was no stranger to such pressure. Nervousness was nonexistent.

After Deng finished speaking, Zhou Yun answered calmly, “This is a topic I want to research. According to my investigation, there is currently no paper titled with this exact topic.”

“Oh? You came up with this topic yourself?” Deng seemed surprised. Even among his doctoral candidates in their third year, none had completely independent research topics.

For a newcomer, this was quite impressive.

“Please continue.”

Zhou Yun flipped to the next slide, which had no text but displayed logos of several renowned AI companies.

“These are some of the most famous AI companies today. GPT, Claude, Gemini—these are among the world’s top large models. Undeniably, these models are extremely powerful, but correspondingly, training them consumes massive resources.

An advanced image recognition model may require tens or even hundreds of megabytes of memory and billions of computational operations.

They are like high-performance sports cars that must run on large cloud data center tracks to perform well. But in real life, our devices are computers, phones, cameras, and so forth.”

Pausing briefly, Zhou Yun flipped to the next slide.

Since the professor hadn’t interrupted, he continued.

“In these circumstances, we face a huge challenge: how to enable these enormous, complex AI models to run quickly and effectively on resource-limited ‘small devices’ while still responding in real time?

Here are four pain points I believe must be addressed:

1. Size mismatch: Large models cannot run on small devices or fit into them.

2. Lagging speed: Even if they barely run, response times are slow, compromising user experience.

3. High energy consumption: Small devices have limited battery life, and large models quickly drain power.

4. Network dependence: Relying on the cloud introduces latency, fails when offline, and raises privacy concerns.”

“Excuse me, let me interrupt.”

“Please, professor.”

“You mentioned that these large models are powerful but resource-intensive, and now you want to cram them into small devices. I understand the difficulty, but since these top-tier models perform so well in the cloud, why insist on forcing them onto edge devices? Isn’t the convenience and scalability of cloud computing more in line with AI’s development trend? How necessary is edge deployment really?”

This was a sharply pointed question. If Zhou Yun couldn’t answer it, the very necessity of his research would be fundamentally undermined.

“Professor, your question is indeed crucial. As you said, cloud-based large models offer unparalleled performance and scalability.

However, large models rely on data!

In highly sensitive industries such as finance, healthcare, and national security, strict laws and regulations require data not to leave the premises or factory.

In such cases, any data transmission to the cloud carries inherent risks. Although strong security measures exist, the cloud itself is a massive ‘target.’

If cloud services suffer a cyberattack, the consequences can be catastrophic. I’m sure you are aware of this.”