Mathematical Methods of Image and Pattern Analysis 2

Please ask any questions that might be of general interest in the feedback forum instead of by mail.

Please go to Moodle for further information and the teaching material.

• Lectures: Prof. Christoph Schnörr
• Exercises: Alexander Zeilmann
• Format:
  • Extensive lecture notes that you read on your own.
  • Short videos that explain each subtopic from a top-down viewpoint. This is not a replacement for studying the lecture notes.
  • Exercise sheets you work on your own and compare with our solution videos
  • Multiple choice tests
  • A feedback forum in Moodle where students can give feedback and ask questions.
  • No meeting in HeiConf, etc.
  • No in-person lecture
• Language: English
• SWS: 4
• ECTS: 6
• Lecture Id: MM35, Spezialisierungsmodul Numerik und Optimierung
• Supplementary Practical: For doing Programming Exercises during the semester you get two extra credits (this might depend on your field of study.)
• Registration: Please register in Müsli and Moodle (you do not need an enrollment key).
• Feedback: We have a feedback forum in Moodle for all questions related to the mathematical content and the organization of the course.
• Prior Knowledge: Required: Lineare Algebra and Analysis, Recommended: Convex and Nonlinear Optimization, Mathematical Methods of Image and Pattern Analysis 1
• Content: This lecture continues and is based on the previous lecture in the summer term 2020. Basic concepts and mathematical background of four interdependent subject areas will be explained:
  • data denoising and restoration;
  • unsupervised learning (clustering, data partitioning without context), data embedding and metric learning;
  • context-sensitive data partitioning and segmentation;
  • correspondence, optimal transport and matching of structures in data.

The intended audience are students of mathematics or scientific computing that are interested in a solid background (and possibly subsequent research) at the intersection of applied mathematics, machine learning and computational data analysis. Some material is motivated by problems of image analysis where an underlying grid graph represents `context'. Concepts and methods still apply after replacing this graph by the binary relation in any other application domain.