Transformer Uncertainty Estimation with Hierarchical Stochastic Attention

Abstract

Transformers are state-of-the-art in a wide range of NLP tasks and have also been applied to many real-world products. Understanding the reliability and certainty of transformer models is crucial for building trustable machine learning applications, e.g., medical diagnosis. Although many recent transformer extensions have been proposed, the study of the uncertainty estimation of transformer models is under-explored. In this work, we propose a novel way to enable transformers to have the capability of uncertainty estimation and, meanwhile, retain the original predictive performance. This is achieved by learning hierarchical stochastic self-attention that attends to values and a set of learnable centroids, respectively. Then new attention heads are formed with a mixture of sampled centroids using the Gumbel-Softmax trick. We theoretically show that the self-attention approximation by sampling from a Gumbel distribution is upper bounded. We empirically evaluate our model on two text classification tasks with both in-domain (ID) and out-of-domain (OOD) datasets. The experimental results demonstrate that our approach: (1) achieves the best predictive-uncertainty trade-off among compared methods; (2) exhibits very competitive (in most cases, better) predictive performance on ID datasets; (3) is on par with Monte Carlo dropout and ensemble methods in uncertainty estimation on OOD datasets.

A Saliency-based Clustering Framework for Identifying Aberrant Predictions

“…A model is perfectly calibrated if for some data distribution D, for all input pairs (x, y) ∈ D, if a model predicts p i = 0.8, then 80% such pairs have y i as a ground truth label. Work on calibration can be broadly categorized into post-hoc methods which calibrate models after training ( [12], [13], [14]), regularization methods during training ( [15], [16], [17], [18]), data augmentation methods ( [19], [20]), and alleviating miscalibration by injecting randomness with uncertainty estimation ( [21], [22], [23]). However, even a perfectly-calibrated model can make aberrant predictions.…”

Deep Learning for Edge Computing Applications: A State-of-the-Art Survey

“…This line of work aims to alleviate model miscalibration by injecting randomness.The popular methods are (1) Bayesian neural networks (Blundell et al 2015;Fortunato, Blundell, and Vinyals 2017), ( 2) ensembles (Lakshminarayanan, Pritzel, and Blundell 2017), (3) Monte Carlo(MC) dropout (Gal and Ghahramani 2016) and ( 4) Gumbel-softmax (Jang, Gu, and Poole 2017) based approaches (Wang, Lawrence, and Niepert 2021;Pei, Wang, and Szarvas 2022). The former three sub-categorgies have been discussed in recent surveys (Mena, Pujol, and Vitria 2021;Gawlikowski et al 2021) ).…”