Enhancing In-context Learning via Linear Probe Calibration

Momin Abbas, Yi Zhou, Parikshit Ram, Nathalie Baracaldo, Horst Samulowitz, Theodoros Salonidis, Tianyi Chen
Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, PMLR 238:307-315, 2024.

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.

Cite this Paper

BibTeX
@InProceedings{pmlr-v238-abbas24a, title = { Enhancing In-context Learning via Linear Probe Calibration }, author = {Abbas, Momin and Zhou, Yi and Ram, Parikshit and Baracaldo, Nathalie and Samulowitz, Horst and Salonidis, Theodoros and Chen, Tianyi}, booktitle = {Proceedings of The 27th International Conference on Artificial Intelligence and Statistics}, pages = {307--315}, year = {2024}, editor = {Dasgupta, Sanjoy and Mandt, Stephan and Li, Yingzhen}, volume = {238}, series = {Proceedings of Machine Learning Research}, month = {02--04 May}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v238/abbas24a/abbas24a.pdf}, url = {https://proceedings.mlr.press/v238/abbas24a.html}, 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. } }
Endnote
%0 Conference Paper %T Enhancing In-context Learning via Linear Probe Calibration %A Momin Abbas %A Yi Zhou %A Parikshit Ram %A Nathalie Baracaldo %A Horst Samulowitz %A Theodoros Salonidis %A Tianyi Chen %B Proceedings of The 27th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2024 %E Sanjoy Dasgupta %E Stephan Mandt %E Yingzhen Li %F pmlr-v238-abbas24a %I PMLR %P 307--315 %U https://proceedings.mlr.press/v238/abbas24a.html %V 238 %X 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.
APA
Abbas, M., Zhou, Y., Ram, P., Baracaldo, N., Samulowitz, H., Salonidis, T. & Chen, T.. (2024). Enhancing In-context Learning via Linear Probe Calibration . Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 238:307-315 Available from https://proceedings.mlr.press/v238/abbas24a.html.

Related Material