Statistical Machine Learning, Week 11: Implementing Convolutional Neural Networks (CNNs) using Keras
#convolutionalneuralnetworks #neuralnetworks #tensorflow
Notebook: https://github.com/farhad-pourkamali/...
Convolutional Neural Networks (CNNs) capture spatial features from images, focusing on pixel arrangements and their relationships
Main Types of Layers:
Convolutional Layer: Extracts features using filters
Pooling Layer: Reduces the dimensionality and highlights dominant features
Fully-connected (FC) or Dense Layer: Combines features for final classification or regression
Advantages over Multilayer Perceptrons (MLPs):
Fewer parameters to learn due to shared weights in filters
Translation invariance: Treats all patches of the image uniformly, regardless of their position
Locality: Uses only small neighborhoods of pixels to compute hidden representations, making it computationally efficient