Scalable Collective Inference from Richly Structured Data using Probabilistic Soft Logic (PSL)
summary
A probabilistic programming framework maps weighted logical rules over richly relational data to a convex optimization by treating random variables as continuous-valued 0–1 variables and deriving equivalent interpretations from logical, randomized rounding and soft logic views. It gives a 3/4 rounding guarantee for the discrete solution and uses ADMM parallel optimizer to scale. Experiments on entity resolution, collective classification, link prediction, image completion and knowledge graph construction show comparable or better accuracy than MLNs while cutting runtime from hours or days to seconds or minutes.
AI-written summary · about
chapters (15)
- 0:00Introduction and motivation
- 2:03Data structure challenges
- 4:11Probabilistic soft logic
- 7:51PSL by example
- 14:52Convex optimization mapping
- 16:16Mathematical interpretations
- 21:05Graphical model connections
- 24:38Logical frameworks
- 25:27Hinge loss MRFs
- 26:52Scaling with ADMM
- 29:52Performance evaluations
- 31:58Diverse applications
- 33:42Knowledge graph construction
- 34:50Future opportunities
- 35:18Concluding thoughts and Q&A
related talks
35mScalable structure learning and inference for domain-specific probabilistic prog
23mLise Getoor, Professor, Computer Science, UC Santa Cruz
14mMulti-Language Probabilistic Programming
01h 03mAn Overview of Probabilistic Programming
36mProbabilistic scripts for automating common-sense tasks
01h 02mOut-of-Distribution Generalization as Reasoning: Are LLMs Competitive?
26mSemi-symbolic Inference for Efficient Streaming Probabilistic Programming
06mPitfalls of Full Bayesian Inference in Universal Probabilistic Programming