Skip to main content

Advanced AI

Module code: EEP55C34 · Semester: Semester 2

Resources​

How it's assessed​

Exam: 40% of the module. TCD doesn't list past papers for it.

Modern AI in depth: generative models, self-supervised and reinforcement learning, transformers and graph neural networks, plus explainability, fairness and robustness.

AssessmentWeight
Practical lab exam 130%
Practical lab exam 230%
Final exam (closed book, computer-based)40%

Summarised from the TCD module descriptor, checked 2026-11-09. Details can change from year to year, so check this year's descriptor and Blackboard.

From students who've done it​

No write-ups yet.

Questions by Year​

No past papers for this module are listed on TCD’s past-papers pages. See How it's assessed above.

Topics

  • Generative models: GANs, normalising flows, energy-based models LO1
  • Self-supervised learning LO2
  • Reinforcement learning: MDPs, value-based and policy-gradient methods LO3
  • Transformers: tokenisation, attention, LoRA, distillation LO4
  • Graph neural networks LO5
  • Meta-learning LO6
  • Explainability and fairness LO7–8
  • Privacy-preserving and robust AI LO9–10
Learning outcomes (exam covers LO7–10)
  1. Implement and evaluate GANs, flows and energy-based models
  2. Design self-supervised pipelines (VAEs, SimCLR, pretext tasks)
  3. Solve MDPs with MC, TD, Q-learning, SARSA, DQN, REINFORCE and actor-critic
  4. Build, fine-tune and distil transformers (tokenisers, LoRA, ViT)
  5. Build graph neural networks (GCN, GAT, message passing)
  6. Apply meta-learning (MAML, Reptile, prototypical, siamese)
  7. Explain models with LIME, SHAP and attribution
  8. Measure and reduce bias with fairness metrics
  9. Use differential privacy, secure MPC and federated learning
  10. Generate adversarial examples and defend against them

Summarised from the module descriptor, checked November 2026.