CSE429

---

Deep Learning

3 Credit Hour Course

Intended For Level 4 Term 1 Students

Prerequisite: CSE329

Modern practices in deep neural networks: hidden units, architectural design, back-propagation and automatic differentiation; Regularization: norm penalties, dataset augmentation, noise robustness, early stopping, parameter tying and parameter sharing; Optimization algorithms: adaptive gradient methods, approximate second-order methods; Linear factor models: probabilistic PCA, factor analysis, independent component analysis (ICA), sparse coding, manifold interpretation of PCA; Deep generative models: autoencoders, generative adversarial networks (GAN), variational autoencoder (VAE); Representation learning: transfer learning and domain adaptation, semi-supervised, self-supervised deep learning, contrastive learning; Deep recommender systems: neural collaborative filtering for personalized ranking, deep factorization machines; Deep learning on graphs; Deep reinforcement learning; Bayesian deep learning; Efficient neural networks: sparsity, parameter and compute efficient neural networks; Multi-task and meta learning, multi-modal learning; Energybased models; Interpretability and analysis of deep neural networks; Causality and explainability in deep learning.