🧠 Deep Learning Projects

This repository contains deep learning projects implemented from scratch using Python and NumPy, as well as using the TensorFlow Keras framework for comparison and experimentation.

---

📘 Overview

This project demonstrates how deep learning models can be built both:

1. From scratch — implementing every component manually (forward pass, backpropagation, gradient descent).
2. Using TensorFlow Keras — leveraging a high-level API for efficient model training and experimentation.

---

📁 Project Structure

Notebook	Description
Neural_Net_from_Scratch.ipynb	Implements a fully connected neural network (MLP) from scratch using only NumPy.
Neural_Net_Using_Framework.ipynb	Implements MLP and CNN models using the TensorFlow Keras framework.

---

✨ Features

• Implementation of forward and backward propagation
• Supports multiple (deep) layers
• Activation functions: Sigmoid, ReLU, and Tanh
• Implements Binary Cross-Entropy loss
• Training with gradient descent
• Trained and Evaluated on CIFAR-10(Cats vs Dogs subset)

---

🧠 Model Performance Summary

Here's a summary of the performance on the test set for each model:

🧩 Model	⚙️ Train Accuracy	🧪 Test Accuracy	📊 Remarks
MLP Model (Implemented from Scratch)	85%	60%	⚠️ Overfitting — No convolution or augmentation used
MLP Model	98%	61%	⚠️ Overfitting
Basic CNN Model	93%	73%	✅ Stable baseline
CNN with BN and Dropout (LeNet-5 style)	80%	75%	👍 Better generalization
CNN with Data Augmentation (LeNet-5 style)	80%	76%	📈 Slight improvement
Larger CNN with Data Augmentation	99%	83%	⚠️ Overfitting sign
Transfer Learning with EfficientNetB0	92%	91% 🏆	🌟 Best performance

---

🏁 Conclusion:

• The MLP model, implemented fully from scratch, shows strong training performance but lacks generalization due to absence of regularization and convolutional features.
• The EfficientNetB0 model achieved the highest test accuracy, showing the benefit of transfer learning.

⚙️ Getting Started

🧩 Prerequisites

Make sure you have the following installed:

• Python 3.x
• NumPy
• Jupyter Notebook
• TensorFlow (for framework-based models)

📦 Installation

Clone this repository:

git clone https://github.com/thakur420/Deep-Learning-Projects.git
cd deep-learning-projects