Hello, detectives. This is the Gadget Crime Division.
As always, a letter has been left here today.
The topic for this week is "Data."
In the past, data was simply considered a collection of numbers and characters, but as time has progressed, we've come to realize that the power of data is more potent than words in the billions.
What kind of message might have been hidden in Uncover the Smoking Gun regarding this?
To, R
**The bottom part of this letter contains content that may include spoilers.
The concept of artificial neural networks, the foundation of deep learning, first emerged in 1980.
Recently, as deep learning technology has advanced, AI has become a buzzword everywhere.
This became possible due to hardware development, leading to a dramatic increase in computational capacity and speed.
In other words, computing power has increased, and the graphics card (GPU) has played a key role, ushering in the era of AI in the 2020s.
Why does deep learning require computational power and speed?
It's because of learning and inference.
A well-classified database is used to train deep learning,
and based on this training information, it makes inferences about newly introduced data.
It might sound complex, but it's quite simple with examples.
Although it's now considered a classic, a representative example is MNIST.
The MNIST dataset contains handwritten digits from 0 to 9.
Deep learning learns from this to distinguish human handwriting.
For instance, when writing the number 8,
everyone has their unique style.
Some people write it in one stroke, while others use two circles.
Some connect the circles, while others keep them separate.
Even though the styles vary when you show the paper to people,
most will say it's an 8.
Similar numbers like 0, 3, and 9 might come to mind, but anyone can easily distinguish them.
That's because the number 8 has unique characteristics.
Deep learning learns these characteristics.
When a deep learning model that has learned from 0 to 9 is shown any number,
it identifies which number's characteristics it possesses. This identification is called inference.
Using this concept, it's possible to show numerous photos of cats and dogs and have the system distinguish between them,
or even differentiate the many objects that appear in the view of a self-driving car.
Alongside computing power, what's important is the database.
Quantity isn't everything; quality data is more crucial.
If you're training on numbers and some unrelated data is included, it might cause confusion.
(In some cases, intentionally including such data can enhance performance.)
In Uncover the Smoking Gun, data is an essential theme and appears throughout the story.
The data of Test Subject X is said to maximize the efficiency of deep learning and show no side effects from chip implantation.
In the Human Recycle Project, the goal is to collect human life as data and use it to create human-like robots.
They go even further to collect pure human data, specifically fetal data.
A deep learning model can perform more specific tasks through additional learning after the initial training.
This process is called fine-tuning.
For example, a model that has learned general language can be additionally trained to understand dialects or speech styles.
It can express the characteristics of dialects based on general language.
This idea appears in the gallery case, suggesting that if one discovers the original training data, they could command the robots.
The auction robot has been trained with auction data for famous paintings like Van Gogh's Starry Night and Da Vinci's Mona Lisa.
Gary, knowing this, manipulates the auction robot to use the gallery's funds freely.
The painter robot was trained based on Gary's old work,
and Robin uses special ink only visible to robots in these early works to manipulate the painter robot.
The No Name that appears in the final scene originally had SPERA as its initial training data,
but it was fine-tuned with SPERO's data, becoming what it is today.
One might think SPERA has disappeared, but SPERA's original data, enhanced with SPERO's brutality, has been further trained, completing him.
The more I learn about deep learning, the more it seems to share similarities with humans.
Through the myriad of data coming in through our five senses, we might still be fine-tuning ourselves.
Perhaps sayings like "see and hear good things" or "surround yourself with good people" could be interpreted from this perspective.
Oh, have I become too engrossed? Enough about deep learning, maybe I'll take a walk now...