# Logit-Based Ensemble Distribution Distillation for Robust Autoregressive Sequence Uncertainties

Source: [Logit-Based Ensemble Distribution Distillation for Robust Autoregressive Sequence Uncertainties](<https://www.sota2.com/research/paper/68fa6f521e209131cea4505d>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-26T01:18:28.677Z

Authors: Yassir Fathullah, Guoxuan Xia, Mark Gales

Published: 2023-05-17T17:21:10.000Z

## Abstract

Efficiently and reliably estimating uncertainty is an important objective in deep learning. It is especially pertinent to autoregressive sequence tasks, where training and inference costs are typically very high. However, existing research has predominantly focused on tasks with static data such as image classification. In this work, we investigate Ensemble Distribution Distillation (EDD) applied to large-scale natural language sequence-to-sequence data. EDD aims to compress the superior uncertainty performance of an expensive (teacher) ensemble into a cheaper (student) single model. Importantly, the ability to separate knowledge (epistemic) and data (aleatoric) uncertainty is retained. Existing probability-space approaches to EDD, however, are difficult to scale to large vocabularies. We show, for modern transformer architectures on large-scale translation tasks, that modelling the ensemble logits, instead of softmax probabilities, leads to significantly better students. Moreover, the students surprisingly even outperform Deep Ensembles by up to ~10% AUROC on out-of-distribution detection, whilst matching them at in-distribution translation.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Text Classification on IMDB (test) (Accuracy)](<https://www.sota2.com/research/sota/text-classification-on-imdb-test-accuracy.md>): Text Classification; IMDB (test); Accuracy: 94.21.

- [Text Classification on Pubmed (test)](<https://www.sota2.com/research/sota/text-classification-on-pubmed-test.md>): Text Classification; Pubmed standard (test); Accuracy: 68.02.

- [Text Classification on TREC (test)](<https://www.sota2.com/research/sota/text-classification-on-trec-test.md>): Text Classification; TREC (test); Accuracy: 71.13.

- [Text Classification on AG News (test)](<https://www.sota2.com/research/sota/text-classification-on-ag-news-test.md>): Text Classification; AG News (test); Accuracy: 82.62.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2305.10384>)

[PDF](<https://arxiv.org/pdf/2305.10384.pdf>)
