Sampling in Graphs Alexandr Andoni (Microsoft Research)
Graph compression Why smaller graphs? use less storage space faster algorithms easier visualization
Preserve some structure Cuts approximately Other properties: Distances, (multi-commodity) flows, effective resistances…
Plan 1) Cut sparsifiers 2) More efficient cut sparsifiers 3) Node sparsifiers
Cut sparsifiers
Approach? [Karger’94,’96]:
Concentration
Applying Chernoff bound
Enough?
Smaller size?
Non-uniform sampling [Benczur-Karger’96]
Strong connectivity Connectivity: 5 Strong conn.: 2
Proof of theorem
ii) Cut values are approximated
Iterative sampling
Comments
BREAK
Smaller relaxed cut sparsifiers [A-Krauthgamer-Woodruff’14]:
Motivating example
Proof of theorem
i) Sketch description
ii) Sketch size ???
iii) Estimation
Estimation illustration dense components
iii) Correctness of estimation
Variance
Dense component estimate
Concluding remarks
Open questions