Out-of-Distribution Detection using Synthetic Data Generation

TL;DR

This paper introduces a novel method using Large Language Models to generate synthetic out-of-distribution data, significantly improving OOD detection accuracy without relying on external OOD datasets.

Contribution

The work presents a new approach leveraging LLMs for synthetic OOD data generation, enabling better OOD detection in text classification tasks without external data.

Findings

Achieves zero false positives in some cases

Outperforms baseline methods significantly

Maintains high in-distribution accuracy

Abstract

Distinguishing in- and out-of-distribution (OOD) inputs is crucial for reliable deployment of classification systems. However, OOD data is typically unavailable or difficult to collect, posing a significant challenge for accurate OOD detection. In this work, we present a method that harnesses the generative capabilities of Large Language Models (LLMs) to create high-quality synthetic OOD proxies, eliminating the dependency on any external OOD data source. We study the efficacy of our method on classical text classification tasks such as toxicity detection and sentiment classification as well as classification tasks arising in LLM development and deployment, such as training a reward model for RLHF and detecting misaligned generations. Extensive experiments on nine InD-OOD dataset pairs and various model sizes show that our approach dramatically lowers false positive rates (achieving a perfect zero in some cases) while maintaining high accuracy on in-distribution tasks, outperforming baseline methods by a significant margin.