HPC application in analysis of large scale complex networks
summary
Shows how Networkit on a GPU (tested on Google Colab T4) loads a ~YouTube dataset of about 1.1M nodes and ~3M edges, computes degree distributions (power‑law exponent ≈2), reports max degree ~28,754 and global clustering ≈0.08, and skips full diameter because APSP is costly. Implements degree‑neighbor and clustering‑vs‑degree helpers, runs PLM (Louvain) and label‑propagation community detectors (≈4.4k vs ≈25k communities), compares partitions with mutual information (~0.84 between Louvain runs), and demonstrates that Networkit gives fast, parallel C++ backend access from Python for large graphs.
AI-written summary · about
chapters (10)
- 0:00HPC in network analysis
- 3:17NetworkX limitations
- 14:04Complex network properties
- 22:13Clustering coefficient
- 30:16Network distance and small-world
- 37:50Assortativity index
- 41:01YouTube friendship case study
- 59:27Community detection
- 1:06:40Community partitioning algorithms
- 1:15:55Future directions and NetworkKit
related talks
39mThorny path to the Large Scale Graph Processing
50mBig Data on Smallish Machines
16mAccelerating Distributed Graph Learning by Using Collaborative In-Network...
26mLarge Scale Graph Processing & Machine Learning Algorithms
39mThorny path to the Large Scale Graph Processing
14mP3: Distributed Deep Graph Learning at Scale
44mHow We Store and Analyze a Large Social Graph
30mScalable Graph-Parallel Inference Algorithms and Systems
