CNN Advanced

# Description

Project exploring advanced CNN architectures and regularization techniques to improve convolutional network performance. Implements and compares multiple variants: baseline CNN without regularization, CNN with Dropout (rate 0.3-0.5), CNN with Batch Normalization, CNN with data augmentation (ImageDataGenerator) and CNN combining all techniques. Each variant is trained on CIFAR-10 and evaluated with accuracy, loss and learning curves. Analyzes overfitting effects and how each technique improves generalization.

$ ./deploy.sh $ git clone

# Key features

$ Comparison of 5 CNN architectures with/without regularization

$ Dropout to prevent neuron co-adaptation

$ Batch Normalization to stabilize training

$ Data augmentation: rotation, zoom, shift, flip

$ Learning curve analysis and overfitting detection

$ Early Stopping and learning rate reduction

# Gallery

Desktop view

Mobile view

# Technologies used

Python TensorFlow / Keras