MixCE: Training Autoregressive Language Models by Mixing Forward and Reverse Cross-Entropies

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Zhang, Shiyue, Wu, Shijie, Irsoy, Ozan, Lu, Steven, Bansal, Mohit, Dredze, Mark, Rosenberg, David
Natura: Preprint
Pubblicazione: 2023
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866910459359133696
author Zhang, Shiyue
Wu, Shijie
Irsoy, Ozan
Lu, Steven
Bansal, Mohit
Dredze, Mark
Rosenberg, David
author_facet Zhang, Shiyue
Wu, Shijie
Irsoy, Ozan
Lu, Steven
Bansal, Mohit
Dredze, Mark
Rosenberg, David
contents Autoregressive language models are trained by minimizing the cross-entropy of the model distribution Q relative to the data distribution P -- that is, minimizing the forward cross-entropy, which is equivalent to maximum likelihood estimation (MLE). We have observed that models trained in this way may "over-generalize", in the sense that they produce non-human-like text. Moreover, we believe that reverse cross-entropy, i.e., the cross-entropy of P relative to Q, is a better reflection of how a human would evaluate text generated by a model. Hence, we propose learning with MixCE, an objective that mixes the forward and reverse cross-entropies. We evaluate models trained with this objective on synthetic data settings (where P is known) and real data, and show that the resulting models yield better generated text without complex decoding strategies. Our code and models are publicly available at https://github.com/bloomberg/mixce-acl2023
format Preprint
id arxiv_https___arxiv_org_abs_2305_16958
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle MixCE: Training Autoregressive Language Models by Mixing Forward and Reverse Cross-Entropies
Zhang, Shiyue
Wu, Shijie
Irsoy, Ozan
Lu, Steven
Bansal, Mohit
Dredze, Mark
Rosenberg, David
Computation and Language
Artificial Intelligence
Machine Learning
Autoregressive language models are trained by minimizing the cross-entropy of the model distribution Q relative to the data distribution P -- that is, minimizing the forward cross-entropy, which is equivalent to maximum likelihood estimation (MLE). We have observed that models trained in this way may "over-generalize", in the sense that they produce non-human-like text. Moreover, we believe that reverse cross-entropy, i.e., the cross-entropy of P relative to Q, is a better reflection of how a human would evaluate text generated by a model. Hence, we propose learning with MixCE, an objective that mixes the forward and reverse cross-entropies. We evaluate models trained with this objective on synthetic data settings (where P is known) and real data, and show that the resulting models yield better generated text without complex decoding strategies. Our code and models are publicly available at https://github.com/bloomberg/mixce-acl2023
title MixCE: Training Autoregressive Language Models by Mixing Forward and Reverse Cross-Entropies
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2305.16958