Deep Learning: Data Representations and Neural Networks

This lecture covers data representations, including data heterogeneity, size, noisiness, and attributes. It explains concepts like Bag of Words, histograms, and visual words. It delves into data pre-processing, missing data, and noisy data cleaning. The lecture introduces data normalization and demonstrates exercises on solution techniques. It progresses to discuss imbalanced vs balanced data, feature expansion, kernel methods, and parametric feature expansion. The lecture concludes with an overview of multilayer perceptrons, training techniques like gradient descent and stochastic gradient descent, and standard architectures like LeNet-5, AlexNet, and ResNet.