Reborn in 2010, the AI ​​in my mind is too powerful.

On the first morning without classes, Chen Yiran woke up at 7:30.

This is because my biological clock has been set. After all, I've been using ELI for almost a month and I wake up at this time every day, so my body has adjusted itself.

He turned over and lay there for another twenty minutes, but finding he couldn't fall asleep, he resignedly got up.

He complained to Jasmine, "Why don't you study more and let me sleep in? My road test is at 10 o'clock, why did you get up so early?"

You didn't tell me that yesterday.

"Not smart at all." Chen Yiran said this before brushing her teeth and washing her face, preparing to eat before going to take her road test.

I've had my temporary driver's license for over ten days, and I'll be able to get my official one after passing the test today.

After a hearty meal, I grabbed a dark sweatshirt and jeans, stuffed all my documents into my bag, and drove downstairs to the DMV.

There were quite a few people on Monday morning, and I waited in line for more than twenty minutes.

It was our turn. The examiner was a Latino man in his fifties, with a beer belly, his uniform buttoned up to the collar. I took his learner's permit, flipped through it, and asked, "Let's go, where's your car?"

Chen Yiran led the examiner to the Porsche.

The examiner walked around the car and said with a hint of envy, "The Porsche is really cool."

"Yeah. He really is handsome."

"But who's coming with you? You need someone over 21 with a driver's license to drive with you, you know that, right?"

Chen Yiran looked at the examiner with an embarrassed expression.

The examiner didn't delve into it, shrugged, and got into the passenger seat with the writing board tucked under his arm.

Chen Yiran sat down, fastened his seatbelt, and gestured to the examiner. Then, instructions began to come out one after another.

"Turn right onto Philadelphia Pike."

He recognized this road; he'd passed through it when he came for the written test. It was a two-lane road, lined with those small shops that had been around for thirty years: shoe repair shops, lottery ticket vendors, Dunkin' Donuts. There weren't many cars on the road, but there were many intersections, with stop signs one after another.

Chen Yiran knew this was all a trap; he had to stop at the stop sign, otherwise he would definitely crash.

Then came changing lanes, parking on the side of the road, and finally parallel parking.

To be honest, I'm a little apprehensive. All those advanced driver assistance systems in my past life have spoiled people; parking has become so easy, unlike nowadays where even reversing radars are so vague.

Turn on the right turn signal. Slowly reverse. Turn the steering wheel half a turn. The car body leans inward, but the angle is wrong; it's too close to the right-hand traffic cone.

I quickly steered the car back to its original position, and it slowly came to a stop.

I caught a glimpse of the examiner drawing a checkmark on the board.

Okay, that's good enough.

Next, there's a small road with a speed limit of 30 miles per hour, a few right turns and one left turn, and then you come to a four-way stop sign.

After stopping, I looked around deliberately for a while before driving over.

"The last one, reverse into the parking space."

Two traffic cones were placed to mark the parking spaces. I put the car in reverse, watched the parking sensors, and this time it worked perfectly; I got in on the first try.

The examiner signed his name on the paper and handed the learning permit back.

"Go inside and take your photo. First, go to the counter and pay the $24 processing fee."

Hurry up and get out of the car to join the queue. The DMV photo booth is always the place with the most ridiculously long queues, and this time was no exception. It took a full half hour to get the photo taken, and then another ten minutes to finally get my driver's license.

I glanced at my phone. 11:40. On my way home, I pulled into Subway and bought a sandwich. Subway food isn't exactly delicious, but the ingredients are fresh, so it's unlikely to be bad.

After arriving home, Chen Yiran lay on the sofa and began planning her studies.

Jasmine.

【exist. 】

"I have absolutely no free time now. Could you elaborate on the study plan you mentioned earlier?"

[Continue with CS223A, three or four days should be enough. Then move on to CS161, Algorithms and Data Structures, and after that, you can look at CS107, Computer Organization and Systems. Also, there's a loose end to CS229; Andrew Ng mentioned convolutional neural networks in the last few lectures, only explaining the concept without elaborating. I'd like to fill that in.]

"Convolutional neural network?"

Treating the image as a two-dimensional matrix and using a set of learnable filters to slide across it to extract features is far more effective than flattening the image into a vector and stuffing it into a fully connected network.

Speak in plain language, say something I can understand.

It's something that can recognize images or faces. I want to try creating a framework that, combined with this, can serve as two separate projects.

"What framework?"

[You define the network structure, and it helps you calculate gradients, train the network, and evaluate it—a complete workflow from start to finish. Similar to Theano, but Theano's symbolic computation interface is difficult to use; you have to use a different programming mindset to write the model definition, and the learning curve can take several months. I want to do something simple that can be done in just a few lines of code.]

"sounds good."

[But the framework has a problem—it's only useful for programmers. There are probably only a few thousand people working on this in the entire ML community right now, so I also want to try adding some elements within this framework to create a demo software.]

"Demo software?"

Yes. You upload an image, and the webpage tells you what's in the image. For example, if you upload a picture of a cat, it tells you it's a cat. Something like that.

"I've seen this on the news in my past life; this kind of thing requires training, right?"

The MNIST handwritten digit dataset can be used as the first test. MNIST is a relatively basic dataset, and it is generally considered to be relatively simple in academia, but it is just right to verify the basic functions of the framework.

If the results are satisfactory, further testing can be conducted using CIFAR-10. CIFAR-10 has ten categories, including common animals and vehicles, and is significantly more challenging to identify than MNIST.

Chen Yiran thought for a moment.

"Fine, you decide. Anyway, I don't understand what you mean."

[I'll make you understand; I have another idea later.]

"Tell me, don't be a riddle teller."

[Face recognition: Upload a photo, and the system identifies faces in the photo. The training data uses LFW (Labeled Faces in the Wild), a publicly available dataset released in 2007, containing over 13,000 labeled faces representing more than 5,700 people. The dataset size is not large, and it can be run on a MacBook.]

Chen Yiran sat up a little straighter.

"Facial recognition? In 2010?"

No one has achieved this at the MacBook level yet. However, I feel the technology is sufficient, and we can try to make it happen.

Chen Yiran thought about it and realized it was indeed a good idea. If it could be made, it would definitely attract a lot of attention, and might even make the news, which would be helpful for Chen Yiran in the future, since it's good to become famous as early as possible.

"Will there be any problems with that?"

[It's inevitable that this will spark privacy discussions. However, you've discussed similar issues before when you were working on web scraping. As long as the code is open source and the functionality is transparent, the controversy itself will attract attention. We haven't profited from this, and we've used only open-source data, so there are no problems.]

"Can you do it right now? Or do you need to learn something more?"

[The framework itself takes one to ten days. Model training and optimization will take a few more days. The web demo will take one or two days. In total, it will take three weeks, assuming we push it forward at full speed every day. Other courses can be postponed. CS223A can be studied concurrently, as the computer vision component overlaps with this project; they can be studied simultaneously.]

"Okay, now that I'm free, I'll give you all the time. You can't really go on a round-the-world trip for no reason."

We were chatting casually when my phone vibrated.

AVE number.

Chen Yiran answered the phone and asked immediately, "Has the car arrived in Los Angeles?"

"No, customs procedures are already complete. The goods will be loaded onto the ship this afternoon, and if things go smoothly, we'll send you the bill of lading tonight; if not, it will be tomorrow or the day after."

This is fantastic news! The condition of the letter of credit is the shipping company's ocean bill of lading.

I'm finally going to get my first pot of gold.