Understanding Deep Learning 2400-ZEWW948

Module 1: Introduction & Supervised Learning
Introduces foundational concepts in supervised learning, including the key principles of predictive modeling and classification tasks.
1. Introduction / Supervised Learning: Provides an overview of data science principles, focusing on labeled data and the development of models to predict outcomes.

Module 2: Shallow & Deep Neural Networks
Explores the evolution of neural network architectures, from simple perceptrons to complex deep learning systems.
2. Shallow Neural Networks: Examines the structure and functioning of single-layer and multi-layer perceptrons for basic pattern recognition.
3. Deep Neural Networks: Discusses multilayer architectures and techniques that enable deep learning's success in tackling complex problems.

Module 3: Loss Functions, Fitting & Performance
Focuses on the core processes of training models, measuring their effectiveness, and avoiding overfitting.
4. Loss Functions: Introduces methods to quantify errors between predicted and actual values, forming the basis of model optimization.
5. Fitting Models: Covers approaches for iteratively adjusting models to align with training data.
6. Gradients and Initialization: Explains how gradient-based optimization and proper initialization improve convergence in training.
7. Measuring Performance: Highlights techniques for evaluating model accuracy and reliability using metrics and validation approaches.
8. Regularization: Describes strategies to prevent overfitting by introducing constraints during model training.

Module 4: Advanced Architectures
Delves into cutting-edge neural network architectures and their applications across diverse domains.
9. Convolutional Networks: Explores how convolutional layers process spatial data, revolutionizing image analysis tasks.
10. Residual Networks: Examines skip connections that enable deep networks to train efficiently by mitigating vanishing gradients.
11. Transformers: Discusses architectures built for sequential data, pivotal in natural language processing and large-scale models, such as GPT3.
12. Generative Adversarial Networks (GANs): Covers adversarial training techniques that enable the generation of highly realistic synthetic data.

Module 5: Continuity and Memory in Networks*
Examines specialized networks capable of encoding memory and handling continuity in data structures.
13. Kolmogorov-Arnold Networks / Hopfield Network: Investigates theoretical approaches for function approximation and associative memory models.