Lecture: Machine Learning for Graphs and Sequential Data

This course builds upon the knowledge you gained in the lecture Machine Learning (IN2064). It provides advanced learning principles and covers more complex data domains. Put simply: This course is "Machine Learning 2".

Information

  • We are currently planning to use "flipped classroom" setup. That is, we will upload lecture recordings at the beginning of each week and then hold in-person sessions for answering any questions about the material.
  • Lecture/Exercise: Wednesday and Thursday, 14:00-15:30.
  • We will primarily communicate through Piazza, while using Moodle to distribute the learning material. The password will be made available on Moodle at a later date.
  • Required knowledge: Content of our Machine Learning lecture
  • Please join our course on Piazza via the link posted on Moodle. We use Piazza as a platform for answering questions and handling all organizational matteres. Please avoid sending us e-mails!
  • If you have any problems signing up for the course on TUMOnline or Moodle, please contact jan.schuchardt [at] in.tum.de

Tentative list of topics

  1. Introduction & Advanced ML Principles
    • Machine Learning, Data Mining Process
    • Basic Terminology
    • Variational Inference
    • Deep Generative Models: VAE, Implicit Models, GANs
  2. Robustness
    • Adversarial attacks
    • Adversarial training
    • Exact robustness verification
    • Relaxed robustness certification (Convex relaxation, Lipschitzness, Randomized smoothing)
  3. Sequential Data
    • ML models for text data and temporal data
    • Autoregressive Models
    • HMMs, Kalman Filter
    • Embeddings (e.g. Word2Vec)
    • Neural Networks (e.g. RNN, LSTM)
    • Temporal Point Processes
  4. Graphs & Networks
    • Laws, Patterns
    • (Deep) Generative Models for Graphs
    • Spectral Methods
    • Ranking (e.g., PageRank, HITS)
    • Community Detection
    • Node/Graph Classification
    • Label Propagation
    • Graph Neural Networks
    • (Unsupervised) Node Embeddings
    • Dynamic/temporal graphs