Sparse Neural Network on CIFAR-10 "Efficient Training of Sparse Neural Networks" is a deep learning project that explores Sparse Evolutionary Training (SET) for convolutional neural networks (CNNs) using the CIFAR-10 dataset. The goal is to improve performance and reduce computation by keeping only the most important connections in the network.

🚀 Features 🌱 Sparse Evolutionary Training (SET) Applies SET to reduce the number of connections in the neural network while preserving or improving accuracy.

⚡ Lightweight Architecture Introduces adaptive pruning and regrowth of weights to maintain a sparse yet efficient network during training.

📉 Accuracy & F1 Score Tracking Tracks accuracy, loss, F1-score, and class-wise performance using confusion matrices and classification reports.

📊 Visualization Plots graphs like:

Accuracy vs Sparsity

Confusion Matrix

Class-wise Accuracy Bar Chart

🧠 Transfer Learning Friendly Starts training from a pre-trained base CNN model and applies sparsity on top of it.

🛠️ Use Cases 🚀 Research on model compression & sparsity

⚙️ Deploying efficient models to resource-constrained environments

📈 Comparing full vs sparse networks on standard datasets

🧾 Dataset 📚 CIFAR-10 60,000 images in 10 classes

50,000 training and 10,000 testing images

Used with standard normalization and batch loading