Feed-Forward Latent Domain Adaptation

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Bohdal, Ondrej, Li, Da, Hu, Shell Xu, Hospedales, Timothy
Natura: Preprint
Pubblicazione: 2022
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866909089546633216
author Bohdal, Ondrej
Li, Da
Hu, Shell Xu
Hospedales, Timothy
author_facet Bohdal, Ondrej
Li, Da
Hu, Shell Xu
Hospedales, Timothy
contents We study a new highly-practical problem setting that enables resource-constrained edge devices to adapt a pre-trained model to their local data distributions. Recognizing that device's data are likely to come from multiple latent domains that include a mixture of unlabelled domain-relevant and domain-irrelevant examples, we focus on the comparatively under-studied problem of latent domain adaptation. Considering limitations of edge devices, we aim to only use a pre-trained model and adapt it in a feed-forward way, without using back-propagation and without access to the source data. Modelling these realistic constraints bring us to the novel and practically important problem setting of feed-forward latent domain adaptation. Our solution is to meta-learn a network capable of embedding the mixed-relevance target dataset and dynamically adapting inference for target examples using cross-attention. The resulting framework leads to consistent improvements over strong ERM baselines. We also show that our framework sometimes even improves on the upper bound of domain-supervised adaptation, where only domain-relevant instances are provided for adaptation. This suggests that human annotated domain labels may not always be optimal, and raises the possibility of doing better through automated instance selection.
format Preprint
id arxiv_https___arxiv_org_abs_2207_07624
institution arXiv
publishDate 2022
record_format arxiv
spellingShingle Feed-Forward Latent Domain Adaptation
Bohdal, Ondrej
Li, Da
Hu, Shell Xu
Hospedales, Timothy
Machine Learning
We study a new highly-practical problem setting that enables resource-constrained edge devices to adapt a pre-trained model to their local data distributions. Recognizing that device's data are likely to come from multiple latent domains that include a mixture of unlabelled domain-relevant and domain-irrelevant examples, we focus on the comparatively under-studied problem of latent domain adaptation. Considering limitations of edge devices, we aim to only use a pre-trained model and adapt it in a feed-forward way, without using back-propagation and without access to the source data. Modelling these realistic constraints bring us to the novel and practically important problem setting of feed-forward latent domain adaptation. Our solution is to meta-learn a network capable of embedding the mixed-relevance target dataset and dynamically adapting inference for target examples using cross-attention. The resulting framework leads to consistent improvements over strong ERM baselines. We also show that our framework sometimes even improves on the upper bound of domain-supervised adaptation, where only domain-relevant instances are provided for adaptation. This suggests that human annotated domain labels may not always be optimal, and raises the possibility of doing better through automated instance selection.
title Feed-Forward Latent Domain Adaptation
topic Machine Learning
url https://arxiv.org/abs/2207.07624