Anomaly detection

The course is based on a series of research papers and projects focused on the anomaly detection problem. Students will review the referenced papers before class such that they can be discussed and expanded upon during class. Lab work will involve applying the concepts tought at during course in order to implement various practical tasks.

Organisation, grading and curricula information can be found in the first deck of slides. The full course curricula can be found here.

The main reference of this course is the Outlier Analysis book.

[1] Charu C. Aggarwal Outlier Analysis, Springer, 2017 [ Springer ]

Professors

Course

Laboratory

Prerequisites

Bachelor courses:

  1. Numerical Recipies
    2. [1] Van Loan, Charles F., and G. Golub Matrix computations, Johns Hopkins University Press, 2013 [ JHU Press ]
    [2] Deisenroth, Marc Peter, A. Aldo Faisal, and Cheng Soon Ong. Mathematics for machine learning, Cambridge University Press, 2020 [ PDF ]

Course materials

During course we will mainly work at the whiteboard supported by the following materials and slides. The papers that need to be prepared by students in advance are marked accordingly or announced in class a week in advance.

  1. Introduction to the problem of anomaly detection: general concepts, examples, motivation
  2. Leverage scores for linear regression
  3. Density based: k-NN, LOF
  4. Tree based: Isolation Forest
  5. Statistical algorithms: truncation, LODA
  6. Distance based: OC-SVM, SVDD
  7. Data adaptation: time series
  8. Applications: network throughput analysis
  9. Dimensionality reduction: PCA, robust PCA
  10. Dimensionality reduction: Autoencoder
  11. Data adaptation: graphs
  12. Applications: banking data
  13. LLM-based anomaly detection

Laboratory classes

  1. Introduction: data generation and tools
  2. Basic anomaly detection algorithms
  3. Isolation Forest
  4. OC-SVM
  5. Autoencoders
  6. Graph-based anomaly detection algorithms