Helmholtz AI FFT seminar series #1: </br>Tingying Peng

# Helmholtz AI FFT seminar series #1: </br>Tingying Peng ###### tags: `HelmholtzAI`,`FFT`,`FFT1` [ToC] ## :memo: Seminar details **8 April 2021, 11:00 - 12:00** - Speaker: **Tingying Peng**, Helmholtz AI YIG @ Helmholtz Munich - Title: **Sparse and low rank matrix decomposition and its applications** _Host: Helmholtz AI central unit_ ## Notes - NMF: https://en.wikipedia.org/wiki/Non-negative_matrix_factorization - difference of PCA and robust PCA: usage of L2 (PCA) and L1 (RPCA) norm - robust PCA: http://www.columbia.edu/~jw2966/papers/CLMW11-JACM.pdf - start of (R)PCA: disentangle input data `X` into low-rank part `L` and sparse part `S` - both `L` and `S` have the same shape as `X` - typically, `S` contains deviations from the "typical" image - typically, `L` contains the typical features of an image - when used on images, `L` and `S` contain different aspects of the image - RPCA using ALM runs a lagrangian optimization to produce `L` and `S` (see slides for details) - related fields: tensor decomposition, compressed sensor, optimization - nice video: https://www.youtube.com/watch?v=yDpz0PqULXQ - nice RPCA Matlab package: https://github.com/andrewssobral/lrslibrary ## :question: Questions for the speaker :::info :bulb: Write down any questions or topics you wish to discuss during the seminar ::: > Leave in-line comments! [color=#3b75c6] :arrow_right: Q1 ... ## :question: your feedback ### Share something that you learned or liked :+1: - cool to see the concept in action for real data and how the theory can be applied - nice talk even with the interruption in the beginning ### Share something that you didn't like or would like us to improve :-1: - trying new presentation technology (ipad with ipen etc) is cool, but I suggest to test-drive this technology **before** the presentation -> we did test it before & had no problems - I was lost quite early in the talk, because I lacked the necessary background and somehow missed the purpose of the presented techniques. -> we will try to present more basic talks - if you ask paramount questions like "Who knows PCA?" and you get silence, I suggest to use hackmd pad or other tools to get the majority of answers; because knowledge of PCA is paramount to understand and appreciate robust PCA -> good idea, we will try to do it next time - where was the networking you talked about in the beginning? -> Sorry, this seminar is not meant for networking per se, but to learn about different research topics and discuss them at the end (due to the problems in the beginning, the discussion part was very short today). Through the lectures, however, you still get to know different people from Helmholtz AI and their projects. :::info :pushpin: Want to learn more? ➜ [HackMD Tutorials](https://hackmd.io/c/tutorials) :::