Enhancing In-context Learning via Linear Probe Calibration

TL;DR

This paper introduces LinC, a calibration method that improves the reliability and performance of in-context learning with GPT models, especially in low-resource and variable prompt scenarios.

Contribution

LinC is a novel calibration technique that enhances ICL robustness and accuracy with minimal additional data, addressing scalability and stability issues.

Findings

LinC improves GPT ICL performance by up to 21%.

LinC reduces calibration error and increases robustness.

Effective in low-resource and permutation-varying settings.

Abstract

In-context learning (ICL) is a new paradigm for natural language processing that utilizes Generative Pre-trained Transformer (GPT)-like models. This approach uses prompts that include in-context demonstrations to generate the corresponding output for a new query input. However, applying ICL in real cases does not scale with the number of samples, and lacks robustness to different prompt templates and demonstration permutations. In this paper, we first show that GPT-like models using ICL result in unreliable predictions based on a new metric based on Shannon entropy. Then, to solve this problem, we propose a new technique called the Linear Probe Calibration (LinC), a method that calibrates the model's output probabilities, resulting in reliable predictions and improved performance, while requiring only minimal additional samples (as few as five labeled data samples). LinC significantly enhances the ICL test performance of GPT models on various benchmark datasets, with an average improvement of up to 21%, and up to a 50% improvement in some cases, and significantly boosts the performance of PEFT methods, especially in the low resource regime. Moreover, LinC achieves lower expected calibration error, and is highly robust to varying label proportions, prompt templates, and demonstration permutations. Our code is available at \url{https://github.com/mominabbass/LinC}.