Rebirth: Starting from Lighting Up the Tech Tree

Chapter 107 Recruiting People

On the third floor of the School of Information Science and Technology at Huaxia University, Zuo Cheng stood at the end of the corridor, looking at the thesis posters on the wall.

This is the AI ​​lab's results display area, where top conference papers from the last three years are printed and posted on both sides of the corridor. Zuo Cheng looked through them one by one, his gaze lingering on a poster.

"An Asynchronous Federated Learning Framework for Resource-Constrained Devices", by Shen Yiming, supervised by Professor Lin Zhiyuan.

Zuo Cheng looked at it for a few seconds and memorized the name.

He pushed open the door to the AI ​​lab. Lin Zhiyuan was sitting behind his desk. When he saw Zuo Cheng come in, he took off his glasses and wiped them.

"Zuo Cheng? What brings you here?"

"Professor Lin, I have something I'd like to discuss with you." Zuo Cheng sat down opposite him. "402 is preparing to establish an AI business unit and needs to recruit a group of AI professionals. We'd like to see if the university has any suitable graduates."

Lin Zhiyuan put his glasses back on and leaned back in his chair: "You're quick on the uptake. Sky Dome Phase III was just delivered, and you're already switching to AI?"

"It's not a shift in focus, it's an expansion," Zuo Cheng said. "Our IoT platform has over 100,000 terminal devices, and the Sky Dome project is also using deep reinforcement learning for spectrum scheduling. AI isn't a new field for us; it's a technological upgrade of our existing business."

Lin Zhiyuan nodded: "That makes sense. But AI talent is in high demand right now, with big companies offering annual salaries of seven or eight hundred thousand yuan. Can your small company compete?"

"Big companies have their problems," Zuo Cheng said. "A PhD graduate who joins a big company is most likely to be a cog in the machine, responsible for a small module within a large team. But in 402, they can directly lead an entire direction, participating in the entire process from algorithm design to engineering implementation. For someone who truly wants to do something, this kind of opportunity is more attractive than salary."

Lin Zhiyuan remained silent for a while, then took out a stack of resumes from the drawer.

Tell me what kind of person you're looking for.

"One to two people each for federated learning, model compression, and computer vision. PhDs preferred, but master's degree holders with strong engineering skills are also welcome."

Lin Zhiyuan flipped through the resumes, took one out, and handed it to Zuo Cheng.

"Take a look at this. Shen Yiming, he's my PhD student who graduated this year. His dissertation focused on federated learning. He has a solid theoretical foundation and strong engineering skills; he even built his own distributed training framework."

Zuo Cheng took the resume, which had a passport photo printed on it. It showed a young man wearing black-rimmed glasses with a somewhat wooden expression.

Where does he work?

"He's unemployed." Lin Zhiyuan sighed. "He interviewed at three major companies, but didn't get the job. The reason is his poor communication skills; he was so nervous during the interviews that he couldn't speak. But in terms of technical ability, he's in the top three of the doctoral students I've supervised over the years."

Zuo Cheng looked at the contact information on the resume, folded it up, and put it in his pocket.

"Can I see him?"

"I'll arrange it for you. He's still in school, and his rented apartment isn't far from here."

Zuo Cheng stood up, then turned back as he walked to the door: "Professor Lin, besides Shen Yiming, if there are any other suitable graduates, please keep an eye out for them. We need people in model compression and computer vision."

"No problem." Lin Zhiyuan waved his hand. "402 is in high demand right now; all the students want to go there."

Zuo Cheng nodded, pushed open the door, and walked out.

Two hours later, Zuo Cheng met Shen Yiming at a coffee shop.

The man was thinner than in the photo, and behind his black-rimmed glasses were a pair of slightly uneasy eyes. He sat in a corner, with an untouched Americano in front of him, his fingers constantly rubbing the rim of the paper cup.

"Shen Yiming?" Zuo Cheng sat down opposite him. "I'm Zuo Cheng, from 402 Technology. Professor Lin showed me your resume, and I'm very interested in your federated learning framework."

Shen Yiming looked up, his lips moved, and after a few seconds he finally spoke: "Hello."

Zuo Cheng wasn't in a hurry. He ordered a coffee and then pointed to the printed copy of the paper in front of Shen Yiming: "I've read your paper on the asynchronous federated learning framework. Your ideas on gradient compression and communication efficiency optimization are ingenious. But I have a question: has the adaptive compression rate adjustment algorithm you proposed in the paper been verified on a real device?"

Shen Yiming's eyes suddenly lit up.

"It's been verified." His speech quickened. "It was done on twenty Raspberry Pis, with a compression ratio dynamically adjusted from 16 to 256 times, communication overhead reduced by 73%, and accuracy loss within 1.5%. But the computing power is too weak for large-scale implementation; the GPU scheduling wasn't approved."

Zuo Cheng listened attentively, his hand secretly manipulating the system panel under the desktop.

Technology radar activated.

The scanning range was thirty meters, and Shen Yiming was only two meters away. The scan results popped up on the panel.

The target was found to possess the following technologies: asynchronous federated learning framework, gradient compression optimization, adaptive communication scheduling, distributed model aggregation, and edge device collaborative training.

Zuo Cheng's pupils contracted slightly.

All five technologies are related to federated learning. Gradient compression optimization and adaptive communication scheduling are precisely the capabilities that 402 architecture engineers need most for edge AI.

Copying requires points, and scanning has already deducted five points. Zuo Cheng decided not to rush into copying and to memorize the information first.

"Your research area aligns perfectly with what we at 402 are doing," Zuo Cheng said. "We have an IoT platform with over 100,000 terminal devices. At 402, you can validate your framework on real industrial equipment without having to wait in line for GPU clusters."

Shen Yiming was taken aback: "More than 100,000 units?"

"Yes, and it's still growing. These devices generate data every day, and your federated learning framework can perform distributed training on these devices, solving data privacy issues while leveraging edge computing power."

Shen Yiming was silent for a moment, then rubbed his fingers together: "But your company isn't very big, is it? The big companies I interviewed with before said that small companies don't have enough data and computing power to support AI research."

"They're right, but 402 is different." Zuo Cheng took out his tablet and opened the monitoring page of the IoT platform. "Four cities, 1,500 edge gateways, 100,000 terminal nodes. Each one is a potential federated learning node."

Shen Yiming stared at the screen, his eyes growing brighter and brighter.

Zuo Cheng closed his tablet: "402 doesn't require you to be eloquent during the interview. We only require you to excel technically. If you're willing, come to the company tomorrow and then make a decision."

Shen Yiming lowered his head and sat quietly for more than ten seconds. Then he raised his head and nodded vigorously: "Okay."

As Zuo Cheng stepped out of the coffee shop, he opened the system panel. The Technology Radar had entered its 48-hour cooldown period, and the score displayed was 290 points.

He noted down the list of technologies Shen Yiming possessed, especially gradient compression optimization and adaptive communication scheduling. If Shen Yiming joined 402, these technologies could naturally be used by the company without needing to spend extra points to copy them.

It's more cost-effective to leverage the skills that talent brings with them than to replicate them using points.

Zuo Cheng walked towards the car parked by the roadside. A core talent in the Federation's learning direction, plus a technology list scanned by the technology radar.

The AI ​​division of 402 has found its first piece of the puzzle.