Regarding this issue, it wasn't just Zhao Wenyuan who noticed it; Meng Yao and Zhan Xiang had also noticed it long ago.
As the founder of Mengnao Technology and the core creator of the entire Mengnao system, Meng Yao certainly understands the principles of how brainwave signals are collected much better than Zhao Wenyuan.
Over the past few years of entrepreneurship, Meng Yao has personally conducted many rounds of experiments, and she has become very proficient in performing brain visualization and stimulating corresponding brain waves.
From any perspective, Meng Yao's test results should have been better than Zhao Wenyuan's, who was a complete novice.
But everyone saw the result just now.
Meng Yao only tried two movements, one of which failed; while Zhao Wenyuan performed almost every movement that had been used for EEG collection in just over ten minutes, without failing any of them.
In terms of latency, although no professional instruments were used for precise measurement, the perceived latency for both individuals was approximately one second.
This time could be the time it takes for the current model to perform inference, or the time it takes for the software driver to map the decoded signals into actions. We haven't done a detailed analysis yet, so it's hard to say.
In conclusion, Zhao Wenyuan's test results were indeed better than Meng Yao's.
Why?
Meng Yao and Zhan Xiang exchanged a glance, already having a guess in their minds. "Could it be because—" Meng Yao hesitated before speaking, "... because I didn't make a move?"
"No movement?" Zhao Wenyuan repeated, realizing the problem. "You mean, my right hand really moved just now, while you relied purely on imagination during the collection and experiment, without actually making any movements? Is there such a big difference between the two?"
The more Meng Yao thought about it, the more likely it seemed, and she said, "It's really possible that there could be such a big difference."
Then she explained to Zhao Wenyuan in detail, "Actually, in our field, imagining doing something but not actually doing it, and doing it with all your might, are two completely different concepts. The former, which is what I just did, we call motor imagination, or MI for short; while the latter, which is what you just did, is called motor execution, or ME for short."
"When we conducted experiments on healthy individuals, we mostly tried the mi method, mainly due to considerations of objective experimental conditions. Because we need to test a wide variety of types of movement, it's unlikely that subjects would be running around while being tested on the Brain-Computer Interface."
"But your situation was different just now. This was your first or second time conducting this kind of experiment, so the instructions Old Zhan gave you skipped the mi guidance and went straight to me. Theoretically, this should bring a stronger and more stable EEG signal, but based on our past experience, the improvement is only a few percentage points. The difference won't be that significant, unless—"
Meng Yao paused here, turning her gaze to Zhan Xiang, as if wanting to confirm whether he shared the same thoughts as her.
After all, Zhan Xiang is the algorithm expert in the Mengnao team.
Sure enough, Zhan Xiang, picking up where Meng Yao left off. "Unless, President Zhao, the solution you provided—combining the basic model with the personal model—is particularly sensitive to this difference between 'me' and 'mi'."
As he spoke, Zhan Xiang opened the file he had just collected on his computer and placed Meng Yao's "clenched fist" and Zhao Wenyuan's "clenched fist" side by side.
The moment Zhao Wenyuan saw the image, he couldn't help but think to himself: Is this what you call a "slight" difference? Where's the "slight" difference?
To be honest, these two waveforms look almost identical side by side. If it weren't for the fact that both files are marked with a clenched fist, it would be hard to believe they represent the same action. However, it is quite obvious that Zhao Wenyuan's file has a larger waveform amplitude and is less smooth, with more jagged edges.
After listening to their analysis, Zhao Wenyuan offered his own guess. "This might be due to the different training methods. The traditional method you used before involved extensive noise reduction and cleaning before manual classification. In the cleaned signal, the difference between mi and me might not be significant. However, in the original data, besides the identified signals, the noise that was originally removed also carried information?"
Zhao Wenyuan proposed this hypothesis because such examples have been commonplace in the field of machine learning.
For example, when doing image recognition, researchers initially design a set of rules to extract edge features, texture features, and other similar elements from the image before training the model. Which features are useful depends entirely on the researchers' preferences and luck.
However, after switching to deep learning, people changed their approach. They didn't care which features were useful; they just put them all into the neural network to learn first, and the recognition effect improved significantly.
The characteristics that people select in advance are ultimately limited by their experience, and sometimes they become a limitation.
After listening, Zhan Xiang nodded in agreement.
The field of the brain is too complex. How neuronal firing leads to complex thinking abilities is something that humanity has not yet fully understood. Brainwaves, which are wave signals derived from neural electrical signals, are even more difficult to analyze.
Therefore, the noise that was previously removed may not have been entirely noise; that's entirely possible.
Now, Zhan Xiang is completely convinced by the machine learning technology approach brought by Zhao Wenyuan.
Even comparing this seventeen minutes to the previous fastest Samsung weeks, how many times greater is that difference?
He couldn't keep up with the calculations, and he didn't want to bother anymore; it was almost over anyway. Zhan Xiang he could start researching machine learning algorithms right now.
"But will this hinder your subsequent research? If the effect is better because I performed the movements, what about the disabled people who can't perform the movements in the first place?" Zhao Wenyuan asked.
Meng Yao smiled after hearing this and said to Zhao Wenyuan, "President Zhao, you don't need to worry about that. For people with disabilities, even if their limbs have lost their mobility or they have unfortunately lost a limb, their brains can still give the instruction to 'perform this action,' but they may be unable to do so due to objective conditions."
"This situation is called an MTR attempt, or MA for short. According to our academic research, the intensity of MA is generally close to ME and stronger than mi. Therefore, we will have to wait until the cooperating subjects conduct the experiment to test how effective this approach is for people with disabilities."
"But in any case, what you brought today has been a great help!"
After this brief experiment, the core executives of Mengnao Technology were noticeably more enthusiastic towards Zhao Wenyuan. After all, he was a lucky charm who solved a major problem that had plagued the team for years; who wouldn't be enthusiastic?
Zhao Wenyuan glanced at his watch; the next meeting was about to start, so he got up to say goodbye.
Before leaving, he asked Pang De about the current production status of the helmet.
Pang De shook his head and said, "Our technology wasn't mature enough before; we hadn't reached the mass production stage. I've been researching Chip."
Zhao Wenyuan understood immediately.
Brain-Computer Interface helmets are also electronic devices, Chip are their core components. However, compared to Large Model that often require high-end Chip costing tens of thousands of yuan, if the helmet only needs to run some small edge models in the future, the requirements for Chip should not be too high.
After all, helmets are large, so larger Chip can be used; there's no need to pursue the ultimate space like with mobile phones. "So, how many of these helmets do we have in total?" Zhao Wenyuan asked.
Meng Yao, who was listening to their conversation, already understood what Zhao Wenyuan meant.
President Zhao has taken a liking to this helmet and wants to take one home to play with.
She smiled and said, "Actually, we don't always use helmets in our experiments. This is just for convenience. The traditional method of using flexible electrode sheets and conductive liquid is the mainstream. We made a batch before, and we have two left over. Why don't you take them back and study them, President Zhao? We can exchange ideas more."
Zhao Wenyuan quickly waved his hand and said, "That's too kind of you."
Just as Meng Yao was about to try to persuade him again, Zhao Wenyuan suddenly added, "By the way, do you have any extra robotic arms?"
Meng Yao was taken aback for a moment, then smiled and said, "Yes, that one also has spare parts."
So Zhao Wenyuan went down to the eighth floor empty-handed, but when he returned, he was carrying a cardboard box containing Mengnao Technology's latest Mark-Tit helmet and a high-tech silver alloy robotic arm.
A great harvest! A great harvest!