Lise Getoor, Professor, Computer Science, UC Santa Cruz
summary
Argues that big data is intrinsically graph-structured and frames three concrete microtasks—collective classification, link prediction, and entity resolution—with specific needs: relational feature construction, joint probabilistic inference, and blocking for scalability. Presents a working tool, Probabilistic Soft Logic, that compiles logic rules into convex losses, solves them by direct optimization (distributed ADMM), and reports much faster performance than discrete MRF sampling; open source and applied to multiple real datasets.
AI-written summary · about
chapters (6)
- 0:00Introduction to graph data
- 3:35Micro-level graph analysis
- 9:40Feature engineering
- 12:58Collective reasoning
- 17:52Scaling graph analysis
- 19:26Probabilistic soft logic
related talks
41mScalable Collective Inference from Richly Structured Data using Probabilistic Soft Logic (PSL)
06mCloud-Native GraphRAG: AI’s Logical Edge
41mEnhancing LLMs with Graph Technology
49mPedro Domingos, Professor, University of Washington
44mBuilding Knowledge Graphs
48mBig Graph: Big Data applied to giant, dynamic graphs
01h 31mBig Data: The Computation/Statistics Interface
01h 02mOut-of-Distribution Generalization as Reasoning: Are LLMs Competitive?