Chapter 50: Falsehood and Truth - Sitting on someone's lap discussing research
On Monday morning, the alarm clock rang, and Xu Aida groggily hummed and snuggled into Liang Siyu's arms.
Her eyelids were so heavy she could barely open them, her whole body ached, she didn't want to go to work, didn't want to get out of bed, she just wanted to go back to sleep.
"Go to sleep." Liang Siyu kissed her cheek and tucked a side-lying pillow into her arms.
The algorithm is quite stable now, and it's about the same whether he or Cohen goes, so let her rest for a day.
She let out a muffled groan and went back to sleep.
He gently kissed her hair, pulled the blanket up, set his alarm for 10:30, and after going downstairs, he first asked the chef to prepare brunch for her before taking his own breakfast.
Cohen swallowed the last sip of coffee and asked, "Is Ada not feeling well?"
Liang Siyu touched his nose: "A little bit, we can go today."
Cohen rolled his eyes inwardly as he looked at his friend's expression.
Good heavens, he really didn't want to guess what had happened, but Ned was completely "different" from before. If Ada were really unwell, he wouldn't be like this.
After the afternoon experiment, Cohen put his arm around Danny's shoulder and invited him to dinner.
He said goodbye to Liang Siyu: "I'll be back later. Please give my regards to Ada."
He didn't want to be the one being shown up by his two best friends.
Unfortunately, Xu Aida was looking forward to his return. She had originally planned for the three of them to discuss the follow-up research plan together, but it would be good to first find out Liang Siyu's thoughts.
After dinner, she suggested, "Ned, let's discuss the plans for the new semester?"
Liang Siyu poured two glasses of iced coconut water and sat down opposite her.
Xu Aida got straight to the point: "Next semester, we can switch the hardware to sEMG, which will provide higher data accuracy. I will confirm this with Cohen again."
"Regarding the effects of the drugs, I'd like to try Adderall first, as it's easier to recruit participants. What do you think?"
Liang Siyu, however, was more interested in trazodone.
This drug has sleep-aiding, analgesic, and mood-improving effects, and is used more frequently in amputee patients, making it significantly more valuable for research.
He sensed something was off again; she seemed genuinely anxious. Last semester, she hadn't fully recovered from the flu, yet she was already rushing to catch up on data.
He understood her research tastes; she could definitely see that trazodone was more worthwhile for research, but she only wanted to choose Addera for the sake of speed.
What was her motivation? She wasn't the kind of person who sought fleeting fame or short-term gains.
“Ada,” he hesitated for a moment, then took her hand, “I have a feeling you seem a bit anxious this year. What are you worried about?”
"Or perhaps I'm mistaken? Anyway, would you like to talk to me?"
Xu Aida's feelings were mixed. He was always meticulous and knew her preferences well, so he could always notice something in the smallest details. Fortunately, he would never have imagined such an outrageous truth as rebirth.
Her lips twitched slightly, but in this matter, she wasn't just looking for "quick results," but had other considerations, methods she had never used in her previous life.
She readily admitted, "Yes, Addera's research value is indeed not as high as that of trazodone. However, I would like to take this opportunity to verify a new algorithm to address the problem of data scarcity at its root."
"Are you planning to complete a methodology paper at the same time? And use this dataset for validation?"
Liang Siyu immediately understood her meaning. She wanted to kill two birds with one stone, using one dataset to achieve two purposes: publishing two papers of different natures.
However, he also had some questions: "Wait, let me think. RNN transfer learning may not be suitable for other drugs, right? This algorithm doesn't have universality, does it? Where is the innovation in the method?"
Xu Aida shook his head: "It's not what you think. It's a more radical improvement. I want to use GANs (Giant Adversarial Networks) to train a generator for electromyography data."
She slowed her speech, "I want to prove that a small amount of real data plus a large amount of generated data can train an effective algorithm."
"What?!" Liang Siyu was dumbfounded.
It took him more than ten seconds to process what he said, and his brows slowly furrowed.
No, this isn't radical, it's subversive. In his view, it completely contradicts the fundamental principles of empirical research.
Hsu Ai-ta tried to explain that GANs (Adversarial Networks) require training two AI models that are adversaries to each other.
One type is a generator, which generates an Adelaide-affected signal based on a regular SEM number; the other type is a discriminator, which distinguishes whether the signal is generated by a forger or is a real signal.
The two models attack and defend against each other, and after multiple iterations, the data generated by the generator reaches a level that is "indistinguishable from the real thing"—making it difficult for the discriminator to identify its authenticity.
At this point, a small amount of real data plus a large amount of generated data can be used to train the algorithm, thus solving the problem of insufficient training data and difficulty in data collection for special participants from the root.
“I understand the idea,” Liang Siyu said hesitantly, “but it’s a bit like a numbers game.”
"Using virtually generated data to train motion control algorithms? How do you ensure that they are consistent with real data?"
He pointed out his concerns: "The discriminator's inability to distinguish may not be due to the virtual data being sufficiently realistic, but rather because your discriminator is not smart enough?"
Xu Aida nodded: "So, we still need to recruit Adela users and collect real data. Just think of my GANs algorithm as hitching a ride; if it doesn't work out, there's no loss, right?"
"If, in the future, the algorithm trained on the virtual data generated by it is just as good as the one trained on the real data, can you admit that my method is effective?"
Liang Siyu sighed. Judging from her words, she planned to complete it privately before reporting to her mentor, which showed that she knew the success rate was not high.
This is a new algorithm that was only published a year ago. What if there are problems with the algorithm itself? Besides, even if she completes the algorithm, she still has to submit it for publication. What about the reviewers? He felt that this approach was really risky.
He advised, "Ada, you also need to think about the submission process. Reviewers might find the methodology insufficiently rigorous. At least, as far as I know, no one in medical journals will accept falsified data."
She felt a little uncomfortable and raised her voice slightly: "That's because you are both arrogant and ignorant, unwilling to look at the progress in other fields."
Why is she so stubborn? He rubbed his forehead: "Is this how you're going to reply to the reviewer?"
"Alright, alright, I'll just say it in private." She admitted that she was just being sarcastic.
Actually, she doesn't plan to submit to medical journals; this is more suitable for comprehensive journals like Nature and PNAS. However, since it involves drugs and medicine, she might still encounter reviewers with a medical background.
Her eyes darted around, and she smiled mysteriously, like a little fox. She simply walked around the table and sat on his lap.
“Ned, can you think of a way to deal with the reviewers?” She hugged his neck and shook him gently.
"You wish!" Liang Siyu sighed and helped her sit up straighter. This little rascal always liked to dump these kinds of explanatory writing assignments on him.
He tapped her forehead. "A reviewer, huh? Think about how to convince Professor Brooke first."
“It’s feasible,” she said matter-of-factly. “It’s easy to find participants for Adderall, but a clean trazodone control group isn’t so easy to find, is it?”
Based on research ethics, they obviously cannot have healthy people take trazodone in order to study motion control algorithms.
"Why isn't there a control group?" Liang Siyu shook his head with a smile. "It's a clean drug effect because patients who take trazodone for primary insomnia."
She paused for a moment, then asked, "Oh, is there a control group?"
She poked his shoulder: "Then, why didn't you say so sooner?"
He shook his head: "Weren't you writing Adela's algorithm? How could I have imagined that you would also try to create something out of nothing, kill two birds with one stone, and solve the problem of trazodone at the same time?"
"What do you mean by fabricating something out of nothing? This is data science!" She glared at him, but then fell silent again.
If there were a way to study trazodone, Professor Brooke would certainly not readily accept the argument that "it's faster to do Addera."
She sighed, preparing to ask if patients with primary insomnia were easy to recruit, but a glance at Liang Siyu's eyes gave her a sudden inspiration.
She hugged his neck tighter again and whispered in his ear, "Ned, Ned, what good reason do I have if I want to be Adela? You must know something, tell me now!"
"Alright, alright." He flinched, his Adam's apple bobbing.
“It is difficult to recruit patients with primary insomnia. It requires a complicated screening process, and perhaps only 5 out of 100 applicants will be suitable.”
This is also why he didn't take the initiative to discuss the new plan with her; he had been thinking about whether there was a better solution.
He sighed and made up his mind: "Since you want to speed things up and test the new algorithm, then let's make Adela."
"As for trazodone, why not just recruit amputees who are taking trazodone directly? Since we're just slowly gathering numbers anyway, we might as well go all the way and study the target population directly."
She leaned in and kissed him: "Don't worry, as long as GANs succeeds, we only need a very small sample size for trazodone, and GANs can mimic their data generation. You won't have to worry about the sample size anymore."
"I'm cautiously optimistic about it," he joked. "Maybe by the time the reviewers reject the paper for the first time, we'll have enough patients."
"Ugh, why did you assume they'd be rejected?" Xu Aida gently nudged his shoulder. Although she knew this method was radical and the submission process probably wouldn't be smooth.
He realized his words were inappropriate, so he tapped the table three times and whispered, "knock the wood."
She laughed: "Why do you medical students always have these mystical little rituals? Where's your scientific spirit?"
Liang Siyu was about to tease him back when, with a "bang," Cohen pushed the door open and came in.
Xu Aida jumped up instantly. Oh my god, that's so weird! Was she just clinging to Ned like a koala?
Cohen didn't expect that he still couldn't dodge it.
He couldn't help but complain, "You two, if you keep going like this, I'm going to anonymously expose you on the forum!"
Xu Aida's face was slightly flushed, but upon hearing his words, she couldn't help but retort: "Then I'll block your post!"
Almost simultaneously, Liang Siyu also said, "Be careful Ada might hack your post."
They looked at each other and burst into laughter.
Cohen didn't expect to be bombarded with "dog food" twice in ten seconds, so he simply turned around and headed for the stairs, forgetting even about the elevator.
Those two are still laughing hysterically. What's so funny? It's just noise pollution!
"Cohen, wait a minute, let's talk about the follow-up research plan!" Liang Siyu suppressed a laugh and called him back loudly.
Follow-up research plans? Do they look like they're discussing research? He's not blind.
Cohen didn't even turn his head: "I have no objection. You two continue your research!"
Inside the restaurant, the two looked at each other and couldn't help but laugh again.
His friend's footsteps had faded into the distance. Liang Siyu looked at the smiling girl and couldn't help but bend down and plant a kiss on her cheek: "Naughty girl."
A flush rose to Xu Aida's face, and fireworks exploded on his chest.
Damn it, I can still vividly recall the scene last night when he called her that in a hoarse voice by the desk.
He grabbed her and practically pushed her into the elevator, through the corridor, and back into the room.
As he opened the door, he asked in a serious tone, "May I?"—as if he were inviting her to a waltz at her graduation prom.
Continue read on readnovelmtl.com
