AI and Machine Learning Projects

This repository contains several machine learning and artificial intelligence projects, each focusing on different techniques and challenges. Below is an overview of each folder and its purpose.

Project Structure

1. Cipher - Solving the Hardest Problem from the Polish AI Olympiad

In this project, Word2Vec is utilized to tackle one of the most challenging tasks from the 1st Polish Artificial Intelligence Olympiad. The approach leverages natural language processing (NLP) techniques to find meaningful vector representations of words, enabling the model to decipher complex patterns and relationships in text data.

2. CIFAR-10 - Training CNN and ResNet18 for Image Classification

This project focuses on training convolutional neural networks (CNNs) on the CIFAR-10 dataset. The implementation includes:

• A simple CNN model.
• A ResNet18 architecture achieving nearly 80% accuracy within approximately 30 minutes of training on a single CPU.

This result is quite efficient given the hardware limitations, demonstrating the effectiveness of deep learning techniques even with constrained computational resources.

3. MCTS - Connect Four with AlphaZero and Standard Monte Carlo Tree Search

This folder contains Python scripts implementing Monte Carlo Tree Search (MCTS) for playing the classic Connect Four game. The implementation includes:

• A traditional MCTS-based agent.
• An AlphaZero-inspired approach leveraging reinforcement learning to improve gameplay decision-making over time.

4. Shakespeare - Text Generation using LSTM and GPT

This project involves generating Shakespearean-style text using deep learning models. The dataset consists of 400,000 Shakespearean lines, and the implementation includes:

• LSTM-based text generation.
• A GPT-inspired model, following the approach introduced by Andrej Karpathy in his famous YouTube lecture.

This project showcases how deep learning can be applied to natural language generation, mimicking the style of historical texts.

5. MNIST - Comparing Machine Learning Libraries for Digit Classification

This project explores different machine learning libraries to determine which one is the most efficient for classifying digits from the MNIST dataset. The following libraries and models are tested:

scikit-learn:

• Decision Tree Classifier
• Random Forest Classifier
• Logistic Regression
• K-nearest neighbors
• Support Vector Machine
• Naive Bayes

TensorFlow:

• Simple Neural Network
• Convolutional Neural Network

PyTorch:

• Simple Neural Network
• Convolutional Neural Network

Additionally, Matplotlib is used for data visualization, and NumPy is employed for general data processing, including building a neural network from scratch.

Getting Started

To explore any of these projects, simply clone this repository:

git clone git@github.com:yourusername/your-repo-name.git
cd your-repo-name

Contributions

If you’d like to contribute to this repository, feel free to open a pull request or raise an issue for discussion.

Acknowledgments

• The Shakespeare GPT implementation follows the method outlined by Andrej Karpathy in his YouTube lecture on GPT from scratch.
• The MCTS and AlphaZero methodology is inspired by reinforcement learning research in game AI.

---

This repository serves as an exploration of different AI methodologies, showcasing applications in NLP, computer vision, and game AI.