Notes on: Ratner, A., Sa, C. D., Wu, S., Selsam, D., & R\'e, Christopher (2016): Data programming: creating large training sets, quickly

Table of Contents

Overview

  • Data programming where users express weak supervision strategies or domain heuristics as "labeling functions", which are programs that lable subsets of data, but that are noisy and may conflict

Notation

  • ratner16_data_progr_031c55aa57bd247de3214605bc78f119d897b407.png denotes a labeling function, and ratner16_data_progr_54ff949188c019fb91dea89a3cf5abd8b1296d73.png the rv. representing output of ratner16_data_progr_031c55aa57bd247de3214605bc78f119d897b407.png
  • ratner16_data_progr_71311614e452e17501211d008bb0f24c29e86d08.png denotes probability ratner16_data_progr_031c55aa57bd247de3214605bc78f119d897b407.png labeling an object
  • ratner16_data_progr_6099d7787556bd805b3b6b41d9bd705c84818945.png denotes probability of a labeling by ratner16_data_progr_031c55aa57bd247de3214605bc78f119d897b407.png being correct