Enregistré dans:
Détails bibliographiques
Auteurs principaux: Chen, Wenlong, Li, Yingzhen
Format: Preprint
Publié: 2023
Sujets:
Accès en ligne:https://arxiv.org/abs/2303.02444
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866916943539208192
author Chen, Wenlong
Li, Yingzhen
author_facet Chen, Wenlong
Li, Yingzhen
contents Transformer models have achieved profound success in prediction tasks in a wide range of applications in natural language processing, speech recognition and computer vision. Extending Transformer's success to safety-critical domains requires calibrated uncertainty estimation which remains under-explored. To address this, we propose Sparse Gaussian Process attention (SGPA), which performs Bayesian inference directly in the output space of multi-head attention blocks (MHAs) in transformer to calibrate its uncertainty. It replaces the scaled dot-product operation with a valid symmetric kernel and uses sparse Gaussian processes (SGP) techniques to approximate the posterior processes of MHA outputs. Empirically, on a suite of prediction tasks on text, images and graphs, SGPA-based Transformers achieve competitive predictive accuracy, while noticeably improving both in-distribution calibration and out-of-distribution robustness and detection.
format Preprint
id arxiv_https___arxiv_org_abs_2303_02444
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Calibrating Transformers via Sparse Gaussian Processes
Chen, Wenlong
Li, Yingzhen
Machine Learning
Transformer models have achieved profound success in prediction tasks in a wide range of applications in natural language processing, speech recognition and computer vision. Extending Transformer's success to safety-critical domains requires calibrated uncertainty estimation which remains under-explored. To address this, we propose Sparse Gaussian Process attention (SGPA), which performs Bayesian inference directly in the output space of multi-head attention blocks (MHAs) in transformer to calibrate its uncertainty. It replaces the scaled dot-product operation with a valid symmetric kernel and uses sparse Gaussian processes (SGP) techniques to approximate the posterior processes of MHA outputs. Empirically, on a suite of prediction tasks on text, images and graphs, SGPA-based Transformers achieve competitive predictive accuracy, while noticeably improving both in-distribution calibration and out-of-distribution robustness and detection.
title Calibrating Transformers via Sparse Gaussian Processes
topic Machine Learning
url https://arxiv.org/abs/2303.02444