Probabilistic Scripts for Automating Common-Sense Tasks

Explore a groundbreaking approach to automating common-sense reasoning tasks in this 36-minute Strange Loop Conference talk. Delve into the world of probabilistic scripting, a declarative programming method that encodes domain knowledge and leverages advanced probabilistic programming techniques. Learn how this innovative approach can be applied to real-world data cleaning tasks, achieving state-of-the-art accuracy on benchmarks. Discover the potential of probabilistic programming in bringing common-sense reasoning to various automation challenges. Gain insights into the Metaprob open-source probabilistic programming language and its applications. Follow along as the speaker demonstrates the process of writing scripts, encoding knowledge, and utilizing probabilistic inference in data cleaning tasks. Understand the preliminary results and the broader implications of this technology for automating complex, common-sense-driven tasks in engineering and beyond.

Introduction
Cleaning messy data
Writing the script
Declarative programming
Data cleaning
Encoding knowledge
Encoding domain knowledge
Pclean
GPT
probabilistic inference
PE architecture
Preliminary results