Learning from Multiple Noisy Partial Labelers

TL;DR

This paper introduces a probabilistic model for programmatic weak supervision that incorporates partial labelers outputting subsets of classes, significantly improving accuracy and scalability in text and image classification tasks.

Contribution

It extends existing frameworks to handle partial labelers, providing a scalable learning algorithm and theoretical guarantees of identifiability.

Findings

Adding partial labels improves text classification accuracy by 8.6 percentage points.

The framework scales to 100k examples in one minute, a 300x speedup.

Achieves zero-shot object classification performance comparable to recent embedding-based methods.

Abstract

Programmatic weak supervision creates models without hand-labeled training data by combining the outputs of heuristic labelers. Existing frameworks make the restrictive assumption that labelers output a single class label. Enabling users to create partial labelers that output subsets of possible class labels would greatly expand the expressivity of programmatic weak supervision. We introduce this capability by defining a probabilistic generative model that can estimate the underlying accuracies of multiple noisy partial labelers without ground truth labels. We show how to scale up learning, for example learning on 100k examples in one minute, a 300x speed up compared to a naive implementation. We also prove that this class of models is generically identifiable up to label swapping under mild conditions. We evaluate our framework on three text classification and six object classification tasks. On text tasks, adding partial labels increases average accuracy by 8.6 percentage points. On image tasks, we show that partial labels allow us to approach some zero-shot object classification problems with programmatic weak supervision by using class attributes as partial labelers. On these tasks, our framework has accuracy comparable to recent embedding-based zero-shot learning methods, while using only pre-trained attribute detectors.