Data Science Fundamentals for Neural Networks

Data Science Fundamentals for Neural Networks

1) Data Science: The Root of Neural Networks

Data is the first and most important part of building a neural network. Without data, a neural network has nothing to learn from. Data is the root of every AI system.

Many people think creating AI only means writing code and creating models, but the reality is different. A neural network first needs high-quality data because the model learns patterns, information, and knowledge from the data we provide.

---

Understanding Data Processing for Neural Networks

When working with text data, beginners often use simple methods from programming languages like Python or JavaScript.

For example, many people use a basic split method to separate words.

However, this approach is not always suitable for training an AI model because it can remove important information from the text.

A good data processing algorithm should convert text into tokens while keeping important characters such as:

• Spaces
• Commas
• Dots
• Slashes
• Symbols
• Other special characters

Keeping these characters is important because they help the AI understand how human language is structured.

If punctuation and symbols are removed before training, the AI may understand words and numbers, but it may not learn how to generate complete sentences with correct punctuation.

---

How to Collect Data for Training a Neural Network?

The first step in creating an AI model is collecting useful data.

There are many sources available where you can collect information. One popular example is online knowledge databases because they contain a large amount of text that can be used for learning and experiments.

For small projects, you can collect data manually.

For larger projects, you can use available APIs and automated methods to collect millions of data samples.

After collecting the data, it should be saved in a structured format so it can be processed easily in the next steps.

---

How to Store Collected Data?

After collecting data, the next step is storing it properly.

You can store data in different file formats, such as:

• Text files
• JSON files
• Binary files

For beginners, JSON is a good option because it is easy to understand and allows data to be stored in an organized structure.

Large AI systems usually use optimized formats because they need faster access to huge amounts of data.

---

Choosing a Programming Language

After preparing the data, you need a programming language to process and train your neural network.

Python is one of the most popular choices because it has many AI and machine learning libraries.

However, if you want to understand how neural networks work internally and build systems from a deeper level, C++ is a powerful choice because it provides more control over memory and performance.

---

Creating Token Data Files

Before training, text data needs to be converted into a format that a neural network can understand.

Create two important files.

First File: Token Information

This file stores the relationship between tokens and their IDs.

Example:

Token ID → Token

Example:

1 → Artificial
2 → intelligence
3 → .

---

Second File: Vector Information

This file stores the numerical representation of tokens.

Example:

Token ID → Vector

A neural network cannot understand words directly, so words must be converted into numbers.

---

Converting Data Into Tokens

The collected text data is converted into smaller parts called tokens.

For example:

Original text:

Artificial intelligence is powerful.

After tokenization:

Artificial
intelligence
is
powerful
.

These tokens are then stored with their IDs so the neural network can process them.

---

Converting Tokens Into Vectors

After creating tokens, the next step is converting them into vectors.

Vectors are numerical representations that allow the neural network to understand relationships between words.

Example:

Artificial → [0.25, 0.71, 0.34 ...]

For small experiments, smaller vector sizes can be used.

For example:

• Small datasets
• Learning projects
• Testing models

can use smaller dimensions.

Large AI models trained on millions or billions of tokens usually use much larger vector sizes.

A separate lecture will explain vectors in more detail.

---

Preparing Data for Neural Network Training

At this point, you have prepared the basic training data:

File 1:

Token ID → Token

File 2:

Token ID → Vector

Now the data is meaningful and ready for the next stage.

However, the neural network is still not complete.

Creating tokens and vectors is only the beginning.

The next important steps include:

• Designing the neural network architecture
• Forward propagation
• Calculating loss
• Backpropagation
• Updating weights
• Training the model using CPU or GPU

These topics will be explained in the next lectures.

---

Conclusion

Building a neural network is a long process, and everything starts with data.

A strong AI system follows this path:

Good Data → Good Tokens → Good Vectors → Good Training → Better AI

In the next lecture, we will explore how vectors work and how neural networks learn from data.