| Category | Difficulty |
|---|---|
| HW | 4 |
18-898D is a special topics course in signal processing: Graph Signal Processing and Geometric Learning taught by professor Jose Moura. The objective of this course is to provide students with toolbox to deal with data supported by graphs. The course shows how to extend traditional DSP toolbox to data supported by graphs (Graph Signal Processing).
- brief review of Linear Algebra concepts
- basics on graphs including graph representations through adjacency and graph Laplacian matrices, graph models, and graph spectral analysis
- Graph Signal Processing covering topics such as graph shift, graph shift invariance, graph signals, graph filtering, graph Fourier transform, graph convolution and modulation, graph frequency and graph spectral analysis of graph signals, among others
- Network science and distributed consensus algorithms
- Geometric Learning to extend deep learning models to learning with data supported by graphs, examples: GCN, GAT, Graph Transformers
The course has 4 to 5 HWs followed by a team projects. The HWs and projects are to be done in a team of 2, which means the HWs are sufficiently lengthy and challenging.
The course doesn't assume any background in signal processing. But it's recommended to have at least some knowledge in classical DSP, graph theory and deep learning to appreciate all the ideas being taught.
The HW's are generally lengthy even for a two man team, so start early. Majority part of HW would be on proving some properties followed by some programming questions. There are no reference books available per se as this is a new topic, research papers will be your best friends for reference and if you are stuck on some problem.