2.8. Artificial Intelligence Mathematics (AIM)

2.8. Artificial Intelligence Mathematics (AIM)

This area covers the advanced mathematical foundations of modern artificial intelligence, with emphasis on deep learning theory, advanced neural architectures, reinforcement learning, and generative models. It provides the theoretical understanding needed to develop novel AI architectures, analyze their properties, and understand their limitations. The focus is on the mathematical principles that enable machines to learn complex representations and make intelligent decisions.

Table 2.8: List of KUs in the Artificial Intelligence Mathematics area.

2.8.1. AIM/Deep Learning Theory  (Core Tier1: 1 hr, Core Tier2: 1 hr) ↑ Back to top

Topics:
Core

  • Universal approximation theorems
  • Optimization landscape of deep networks
  • Generalization in deep learning
  • Representation learning theory
  • Training dynamics and gradient flow

Learning Outcomes:
Core:

  1. Explain universal approximation theorems and their implications [Familiarity]
  2. Analyze the optimization landscape of deep networks [Assessment]
  3. Study generalization phenomena in deep learning [Assessment]
  4. Analyze training dynamics of neural networks [Assessment]

2.8.2. AIM/Advanced Neural Architectures  (Core Tier1: 1 hr, Core Tier2: 1 hr) ↑ Back to top

Topics:
Core

  • Convolutional Neural Networks (CNNs)
  • Recurrent Neural Networks (RNNs) and LSTMs
  • Transformer architectures
  • Attention mechanisms
  • Graph Neural Networks (GNNs)

Learning Outcomes:
Core:

  1. Design CNN architectures for visual tasks [Usage]
  2. Implement RNNs for sequence modeling [Usage]
  3. Apply transformer architectures to various domains [Usage]
  4. Use attention mechanisms in neural networks [Usage]

2.8.3. AIM/Reinforcement Learning  (Core Tier1: 1 hr, Core Tier2: 1 hr) ↑ Back to top

Topics:
Core

  • Markov Decision Processes (MDPs)
  • Value iteration and policy iteration
  • Q-learning and temporal difference learning
  • Policy gradient methods
  • Deep reinforcement learning

Learning Outcomes:
Core:

  1. Formulate problems as Markov Decision Processes [Usage]
  2. Implement value and policy iteration [Usage]
  3. Apply Q-learning for model-free RL [Usage]
  4. Use policy gradient methods [Usage]

2.8.4. AIM/Natural Language Processing with Transformers  (Core Tier1: 1 hr, Core Tier2: 1 hr) ↑ Back to top

Topics:
Core

  • Transformer architecture and self-attention
  • BERT and pre-trained language models
  • GPT models and autoregressive generation
  • Multimodal transformers
  • Efficient transformer variants

Learning Outcomes:
Core:

  1. Implement transformer models from scratch [Usage]
  2. Fine-tune pre-trained language models [Usage]
  3. Apply GPT models for text generation [Usage]
  4. Use multimodal transformers for cross-modal tasks [Usage]

2.8.5. AIM/Generative Models and GANs  (Core Tier1: 1 hr, Core Tier2: 1 hr) ↑ Back to top

Topics:
Core

  • Generative Adversarial Networks (GANs)
  • Variational Autoencoders (VAEs)
  • Normalizing flows
  • Diffusion models
  • Evaluation of generative models

Learning Outcomes:
Core:

  1. Implement GANs for image generation [Usage]
  2. Apply VAEs for latent representation learning [Usage]
  3. Use diffusion models for generation [Usage]
  4. Evaluate generative model quality [Assessment]

Spotted a typo, an outdated course, a broken link, or have a suggestion? Let us know.

Scan to open on your phone