訊號處理之數學方法及演算法(一)

This course is targeted for first-year graduate students (but students in their junior or senior year are strongly encouraged to register) who are interested in data science and the mathematics behind it, namely linear algebra, optimization, probability and statistics, which ultimately leads to deep learning. The course will briefly review basic concepts in linear algebra before discussion on matrix computation algorithms, splitting algorithms, compressed sensing, matrix completion, and graph theory. This will culminate into the topic of deep learning, which includes discussions on convolutional neural network (CNN), backpropagation, graph convolutional neural network (GGN), and topology-adaptive graph neural network (TAGCN). The materials presented are not available in courses such as machine learning or deep learning as these courses do not usually develop a deep connection between linear algebra, optimization theory and deep learning for general data science problems (going beyond computer vision or natural language processing).

Linear algebra, probability, and calculus

The official course website https://mcube.nctu.edu.tw/~cfung/courses/2021_2022/MMASP_I/ contains a detailed syllabus and schedule that the students should reference in order to decide whether or not to take the course Google Meet: https://meet.google.com/kjt-cbeg-oqi

Grading: (Tentative) -- Written and/or programming assignments (20%) -- Due time depending on assignments. Hand in before end of class. -- Each day will result in 50% reduction of the full grade. -- Class project (30%) -- Due date: Jan 5, 2022 at 23:59:59. Email presentation (ppt) and code (Matlab) in zip file to your TA. -- Oral presentation (15 min) per person, Q/A (5 min). Jan. 6, 2022. Time: 13:20-16:20 -- Credit given to students who ask “good” questions. -- Each day will result in 50% reduction of the full grade. -- Class Participation (10% at most) -- Given to students who actively participate in class, for example, answering or asking good questions. -- No credit will be given for simply showing up. -- Given to those whose grades are borderline -- Midterm (25%) -- Length: 2 hour exam. -- Scope: From beginning to matrix completion. -- Handwritten double-sided A4 cheat paper. -- Final (25%) -- Length: 2 hour exam. -- Scope: Comprehensive. -- Handwritten double-sided A4 cheat paper.

-- Textbook: G. Strang, Linear Algebra and Learning From Data, Cambridge Press, 2019. -- References: T.K. Moon and W.C. Stirling, Mathematical Methods and Algorithms for Signal Processing, Prentice Hall, 2000. Despite its breadth, this book contains numerous errors, but the authors have generously offered an errata sheet. E.K.P. Chong and S.H. Zak, An Introduction to Optimization, 4th Ed., Wiley, 2013. I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning, MIT Press, 2016. J. Nocedal and S.J. Wright, Numerical Optimization, 2nd Ed., Springer, 2006. -- Lecture notes.

地點

ED 639

時間

By appointment

聯絡方式

Email: c.fung@ieee.org Telephone: 03-573-1862