Notes

Not related to DimSum, but things that I find interesting

TOC .

1. Robust Spectral Clustering
2. Domain Rules
3. Funda Mentals

Detailed Notes

Robust Spectral Clustering

Paper: link
Group: TMU, Technical University of Munich link .
Key concepts : Works around noise issues in spectral clustering.
2 latent factor - clean, corrupt.
Goal optimize eigenspace of laplacian.
Approach: Model: Sparse latent decomposition, normalized laplacians. Algos: Eigen perturbation, Multi Dimensional Knapsack. Experiments: Local purity and global separation.

Domain Rules

improving learning with domain rules, video , not much progress, modified loss function to bias towards some domain things, alpha beta

Funda Mentals

1. eigenvalues and eigenvectors vector that does not change with scaling, so great generalisation and yet expresses the data well
2. graph laplacian like partial differential equations discretized in the natural way
3. k-means, density-based clustering, Normalized mutual information NMI, silhoutte coefficient .
4. Spectral clustering basics link