🧠 PyTorch Deep Learning Portfolio

Collection of PyTorch deep learning projects covering fundamental and advanced architectures: MLPs, CNNs, RNNs, LSTMs, and Transfer Learning.

---

📚 Projects

Project	Architecture	Dataset	Accuracy	Description
MLP-MNIST	Multi-Layer Perceptron	MNIST	~98%	Handwritten digit classification with a 3-layer MLP
CNN-EfficientNet	EfficientNet-B0 (Transfer Learning)	CIFAR-10	~98%	Image classification using pretrained CNN
LSTM-Sentiment	Bidirectional LSTM + Attention	IMDB	~88%	Sentiment analysis on movie reviews

---

🚀 Quick Start

# Clone the repository
git clone https://github.com/cmhh22/Pytorch-deep-learning.git
cd Pytorch-deep-learning

# Choose a project
cd MLP-MNIST     # or CNN-EfficientNet

# Install dependencies
pip install -r requirements.txt

# Run training
python main.py train

---

📁 Repository Structure

Pytorch-deep-learning/
├── MLP-MNIST/              # 🔢 Multi-Layer Perceptron on MNIST
│   ├── models/
│   ├── src/
│   ├── notebooks/
│   ├── main.py
│   └── README.md
│
├── CNN-EfficientNet/       # 🖼️ Transfer Learning on CIFAR-10
│   ├── models/
│   ├── src/
│   ├── notebooks/
│   ├── main.py
│   └── README.md
│
├── LSTM-Sentiment/         # 🎭 LSTM for Sentiment Analysis
│   ├── models/
│   ├── src/
│   ├── notebooks/
│   ├── main.py
│   └── README.md
│
├── LICENSE
└── README.md               # 📖 This file

---

🎯 Learning Path

1️⃣ Start with MLP-MNIST

• Understand basic neural network concepts
• Learn forward/backward propagation
• Master PyTorch fundamentals

2️⃣ Move to CNN-EfficientNet

• Learn convolutional neural networks
• Understand transfer learning
• Apply pretrained models to new tasks

3️⃣ Learn LSTM-Sentiment ✅

• Understand recurrent neural networks
• Learn sequence modeling with LSTMs
• Implement attention mechanisms for interpretability

4️⃣ Coming Soon

• 🔜 Transformers for NLP
• 🔜 GANs for image generation

---

📊 Results Summary

MLP-MNIST

Dataset: MNIST (60K train, 10K test)
Architecture: 784 → 512 → 256 → 10
Test Accuracy: ~98%
Training Time: ~2 min (GPU)

CNN-EfficientNet

Dataset: CIFAR-10 (50K train, 10K test)
Architecture: EfficientNet-B0 (pretrained ImageNet)
Test Accuracy: ~98%
Training Time: ~10 min (GPU)

LSTM-Sentiment

Dataset: IMDB Movie Reviews (50K reviews)
Architecture: Bidirectional LSTM + Attention
Test Accuracy: ~88%
Training Time: ~15 min (GPU)

---

🛠️ Technologies

• PyTorch - Deep learning framework
• TorchVision - Pretrained models & datasets
• NumPy - Numerical computing
• Matplotlib - Visualization
• Scikit-learn - Metrics & evaluation

---

📚 References

• PyTorch Documentation
• EfficientNet Paper
• MNIST Dataset
• CIFAR-10 Dataset

---

📄 License

MIT License - see LICENSE for details.

---

Created by Carlos Manuel • Part of the Deep Learning learning journey 🚀