4F10: Deep Learning

2022, Jun 01    

A collection of topics and materials for 4F10: Deep Learning and Structured Data in 2022.

Key of what’s expected (hopefully…):

  • 🔴 important topic, mathematical detail
  • 🟡 understanding of concepts

  • 🟢 awareness

  • Lecture 1: “Introduction”
    • Nothing of use presented in this lecture
  • Lecture 2: “Probability of Error & Decision Boundaries”
    • 🔴 Much better explained in this very clear tutorial on Linear and Quadratic Disciminant Analysis: arXiv
    • 🔴 Bayes decision rule: slides 4-6
  • Lecture 3: “Graphical Models and Conditional Independence”, and Undirected GMs (Markov Networks)
    • 🔴 CIs: slides 4-6
    • 🟡 Markov networks: Murphy ch19, section 19.3.1.
  • Lecture 4,5,6: “Latent Variable and Sequence Models”
    • 🟡 Factor Analysis sklearn
    • 🔴 EM for learning GMM - Murphy 11.4.2
    • HMMs:
      • 🔴 General: see 3F8 notes
      • 🟡 Inference: Discrete Kalman Filter & Viterbi algorithm: Murphy section 17.4, see also 3F8
      • 🟡 Learning: EM (not covered really)
    • 🟢 Conditional Random Fields - model description, motivation, and learning Murphy section 19.6
  • Lecture 7,8: “Deep Learning”
    • A very poor intro to DL where everything is just in the wrong order - see instead chapters in the tutorials on d2l.ai:
    • 🔴 Initialisation and Xavier initialisation: deeplearning.ai
  • Lecture 9,10: “Deep Learning for Sequence Data”
  • Lecture 11: “Ensemble Methods”
  • 🔴 Lecture 12: “Support Vector Machines” - ok, this is actually quite good
  • 🔴 Lecture 13: “Support Vector Machines: Advanced Topics” (kernel SVM) - good too
  • 🟢 Lecture 14: “Kernels for Structured Data” - good enough
# cambridge