NEURAL-NETWORKS

KO | EN

gitlite — search

#hacktoberfest #ai #go #golang #machine-learning #docker #windows #awesome #kubernetes #awesome-list #vue #security #ios #open-source

NEURAL-NETWORKS

(★ 43)

Solving little tasks involving using NNs

NEURAL-NETWORKS 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

<h1 align="center">NEURAL NETWORKS</h1>

<h2 align="center" style="font-size:50%;">
  <b>BUILDING NEURAL NETWORKS — FROM STRUCTURE TO INTELLIGENCE</b>
</h2>

<br>

![Neural Networks](https://raw.githubusercontent.com/Dreamerol/Dreamerol/main/NN.webp)

<br>

</a>

</div>

<br>

<h1 align="CENTER">🧠 Neural Networks — How Machines Learn to Think</h1>

<p align="center">
  <span style="font-size:170px;">
    <b><i>A portfolio project exploring how neural networks learn — from geometry to dynamic systems.</i></b>
  </span>
</p>

<p align="center">
  <span style="font-size:170px;">
    <b><i>From structure → to patterns → to learning → to time.</i><b>
  </span>
</p>

---

<h2 align="CENTER">📖 PROJECT STORY / ИСТОРИЯ НА ПРОЕКТА</h2>

<p align="center">
  🌐 Neural networks are often presented as tools — fast, efficient, and ready to use.  
</p>

<p align="center">
  But beneath that simplicity lies something far more interesting: a system that learns gradually, through structure, error, and time.
</p>

<p align="center">
  This project is an exploration of that process.  
</p>

<p align="center">
  Not just what neural networks do — but how they become capable of doing it.
</p>

---

<h3>📍 It begins with space</h3>

With points in 2D and 3D geometry.

At first, there is no meaning — only coordinates.  
But through learning, the model begins to separate structure from noise, discovering regions, relationships, and organization inside raw space.

What starts as scattered data becomes an internal map of understanding. ✨

---

<h3>🌊 Then comes pattern</h3>

From space, the problem shifts to functions.

The network is given continuous signals like sin(x), and must learn to approximate what it has never fully seen.  
Not by memorizing points — but by learning the shape behind them.

This is where generalization begins. ⚙️

---

<h3>🔍 Then comes learning itself</h3>

Backpropagation and gradient descent are implemented from scratch.

No abstraction. No hidden mechanics.

Only small updates, repeated over time — each one reducing error, each one refining the model slightly more than before.

Learning becomes visible as a process of continuous correction. 📉

---

<h3>⏳ Finally, time</h3>

The model is extended into dynamic systems inspired by SIR processes.

Now, prediction is no longer static — it evolves.  
Each state depends on the previous one, forming trajectories instead of isolated outputs.

The system begins to understand change itself. 🌍

---

<h2 align="LEFT">🏆 THE RESULT</h2>

A structured progression:

Geometry → Patterns → Learning → Time

More than a project — a narrative of how intelligence can emerge from simple rules, repeated over time.

<p align="left">
At its core, one idea remains:
</p>

<p align="left"><b>
Learning is not a moment.  
It is a process that unfolds.
</b></p>

</div>

---

## 📌 Key Tasks

### 📈 Neural Network Design with TensorFlow
Used for vector classification (determining in which quadrant or octant a vector is located) and function approximation.

#### 🟢 Vector Classification
- **2D Case (Quadrants):**  
  [Vector Classification Quadrants](https://github.com/Dreamerol/NEURAL-NETWORKS/blob/main/NN_vector_classification_2D_quadrants.py)

- **3D Case (Octants):**  
  [Vector Classification Octants](https://github.com/Dreamerol/NEURAL-NETWORKS/blob/main/NN_vector_classification_3D_octants.py)

#### 🔵 Predicting Function Values
- **Sin(x) Prediction:**  
  [Predicting the Behavior of sin(x)](https://github.com/Dreamerol/NEURAL-NETWORKS/blob/main/NN_predicting_the_behaviour_sinx_func.py)

This model predicts the value of sin(x) using a neural network trained on sampled points.

---

### 🔁 Implementing Backpropagation from Scratch
- Implemented manual backpropagation by calculating derivatives and applying gradient descent.
- Explored different neural network architectures and input formats across multiple tasks.
- Used activation functions such as **identity** and **sigmoid**.
- Optimized weights to minimize error using **Mean Squared Error (MSE)**.

### 🤖 Model Evaluation (Accuracy & Loss Functions)
- [Backpropagation Task One](https://github.com/Dreamerol/NEURAL-NETWORKS/blob/main/backpropagation_task1.py)  
- [Backpropagation Task Two](https://github.com/Dreamerol/NEURAL-NETWORKS/blob/main/backpropagation_task2.py)  
- [Backpropagation Task Three](https://github.com/Dreamerol/NEURAL-NETWORKS/blob/main/backpropagation_task3.py)  
- [Finding Best Parameters (Optimization Function)](https://github.com/Dreamerol/NEURAL-NETWORKS/blob/main/Finding_the_best_parameters_using_minimizing_function.py)

- 💡 **Math Modelling:** Applying neural networks for dynamic systems - using ODE/ordinary different equations/ for computing the affection amongst the different variables - infected, sustainable, and recovered
predicting the trajectory of SIR model - plotting the results - interesting fact for these kind of models - is that for training our NN model we need to map the current value of the variables with the previous ones
[SIR_Model](https://github.com/Dreamerol/NEURAL-NETWORKS/blob/main/NN_SIR_model_predictions.py)

---

## 🛠️ Tech Stack / Инструменти

---

## 🚀 Neural Network Lab: From Scratch to Smart Predictions

Welcome to **Neural Network Lab**, where math, code, and logic collide to create intelligent systems.  
This lab is not just coding – it’s understanding **how machines think** and **how neural networks learn from data**.

---

## 🏆 Overview
The lab is divided into several main areas:

- **🔹 Neural Network Design with TensorFlow** – classifying vectors in 2D & 3D.  
- **🔹 Implementing Neural Networks from Scratch** – mastering gradient descent, backpropagation, and weight optimization.  
- **🔹 Mathematical Modelling & Dynamic Systems** – applying NNs to SIR epidemiological models.

Each section teaches **concepts, problem-solving, and visualization**, making abstract math tangible.

**Badges:**    
`📊 TensorFlow` `🟢 Vector Classification` `⚙️ Backpropagation` `📈 Math Modelling` `💡 Scientific Method`

---

### 4️⃣ Math Modelling & SIR Predictions
- Applied NNs to **dynamic systems**: predicting interactions among infected, recovered, and sustainable populations.  
- Learned to **map current values to previous ones**, enabling **time-based predictions**.  
- Plotted **SIR trajectory** to visualize epidemic evolution.

`📊 SIR Modeller` `🌡️ Epidemic Predictor` `📈 Data Visualizer`

---

<h2 align="center">⭐ Feel free to explore repos and give a star if you find them interesting</h2>

Mihaela Koseva (Михаела Косева) • Sofia University (Софийски университет) • AI Engineer • Software Engineer • Backend Engineer • Data Systems & APIs • Applied Machine Learning • Deep Learning • Neural Networks • Model Training • Data Pipelines • Data Science • LLMs • Python • C++ • Java • Clojure • SQL • PyTorch • TensorFlow • Scikit-learn • Pandas • NumPy • ETL • Data Modeling • MLOps
</p>

<p style="font-size:10px; opacity:0.7;">
© 2026 Mihaela Koseva (Михаела Косева) • Софийски университет • Original portfolio design.
</p>

<p style="font-size:10px; opacity:0.7;">
🔗 Explore on GitHub:
<a href="https://github.com/Dreamerol">Mihaela Koseva (Михаела Косева) • Software Engineer • AI • ML • Dreamerol</a>
</p>

</div>

</div>

---

</tr>
</table>

NEURAL-NETWORKS

NEURAL-NETWORKS 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

NEURAL-NETWORKS

NEURAL-NETWORKS 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README