📘 Deep Learning Models and Representation Learning

This repository is a collection of deep learning models and experiments covering core neural network architectures, representation learning techniques, and sequence modeling methods. The codebase is organized by model family, with each directory focusing on a specific concept or learning paradigm. The implementations are designed for learning, experimentation, and research-oriented exploration.

Most of these implementations originate from coursework and projects for the Deep Learning course at Sharif University of Technology.

---

📂 Repository Structure

CNN (Convolutional Neural Networks)

Experiments and models based on convolutional architectures for feature learning and visual tasks:

• CNN_Triplet_Loss_Classification – Metric learning using triplet loss for robust classification.
• Deformable_Convolution – Adaptive convolutional layers for handling geometric variations.
• Displacement_Detection – CNN-based detection of spatial displacement patterns.

---

FFA (Feature-Field Alignment)

Methods for aligning feature representations across domains or distributions:

• Supervised – Feature alignment using labeled data.
• UnSupervised – Distribution alignment without explicit supervision.

---

Fully_connected

Implementations of classical fully connected (dense) neural networks, serving as baselines and foundations for deeper architectures.

---

LSTM (Long Short-Term Memory Networks)

Sequence modeling and temporal learning tasks:

• FC_LSTM_BiLSTM – Comparison of fully connected, LSTM, and bidirectional LSTM architectures.
• Poetry_Generation – Text generation using recurrent neural networks.

---

PCA (Principal Component Analysis)

Dimensionality reduction and feature extraction using classical linear techniques.

---

VAE (Variational Autoencoders)

Probabilistic generative models for representation learning:

• VAE_CVAE.ipynb – Variational Autoencoder and Conditional VAE implementations.
• VQ_VAE.ipynb – Vector Quantized VAE for discrete latent representations.

---

🎯 Purpose

This repository serves as:

• A learning resource for deep learning fundamentals
• A research playground for experimenting with architectures and loss functions
• A modular reference for CNNs, RNNs, and generative models

Each folder is self-contained and focuses on a specific modeling idea, making the repository easy to extend and maintain.

---

🛠️ Notes

• Most experiments are implemented in Python using common deep learning frameworks.
• Notebooks (.ipynb) include explanations and visualizations where appropriate.
• The repository is actively evolving as new models and ideas are explored.