Semi-supervised learning has proven to be a powerful paradigm for leveraging
unlabeled data to mitigate the reliance on large labeled datasets. In this
work, we unify the current dominant approaches for semi-supervised learning to
produce a new algorithm, MixMatch, that works by guessing low-entropy labels
for data-augmented unlabeled examples and mixing labeled and unlabeled data
using MixUp. We show that MixMatch obtains state-of-the-art results by a large
margin across many datasets and labeled data amounts. For example, on CIFAR-10
with 250 labels, we reduce error rate by a factor of 4 (from 38% to 11%) and by
a factor of 2 on STL-10. We also demonstrate how MixMatch can help achieve a
dramatically better accuracy-privacy trade-off for differential privacy.
Finally, we perform an ablation study to tease apart which components of
MixMatch are most important for its success.
Описание
[1905.02249] MixMatch: A Holistic Approach to Semi-Supervised Learning
%0 Generic
%1 berthelot2019mixmatch
%A Berthelot, David
%A Carlini, Nicholas
%A Goodfellow, Ian
%A Papernot, Nicolas
%A Oliver, Avital
%A Raffel, Colin
%D 2019
%K semi-supervised-learning
%T MixMatch: A Holistic Approach to Semi-Supervised Learning
%U http://arxiv.org/abs/1905.02249
%X Semi-supervised learning has proven to be a powerful paradigm for leveraging
unlabeled data to mitigate the reliance on large labeled datasets. In this
work, we unify the current dominant approaches for semi-supervised learning to
produce a new algorithm, MixMatch, that works by guessing low-entropy labels
for data-augmented unlabeled examples and mixing labeled and unlabeled data
using MixUp. We show that MixMatch obtains state-of-the-art results by a large
margin across many datasets and labeled data amounts. For example, on CIFAR-10
with 250 labels, we reduce error rate by a factor of 4 (from 38% to 11%) and by
a factor of 2 on STL-10. We also demonstrate how MixMatch can help achieve a
dramatically better accuracy-privacy trade-off for differential privacy.
Finally, we perform an ablation study to tease apart which components of
MixMatch are most important for its success.
@misc{berthelot2019mixmatch,
abstract = {Semi-supervised learning has proven to be a powerful paradigm for leveraging
unlabeled data to mitigate the reliance on large labeled datasets. In this
work, we unify the current dominant approaches for semi-supervised learning to
produce a new algorithm, MixMatch, that works by guessing low-entropy labels
for data-augmented unlabeled examples and mixing labeled and unlabeled data
using MixUp. We show that MixMatch obtains state-of-the-art results by a large
margin across many datasets and labeled data amounts. For example, on CIFAR-10
with 250 labels, we reduce error rate by a factor of 4 (from 38% to 11%) and by
a factor of 2 on STL-10. We also demonstrate how MixMatch can help achieve a
dramatically better accuracy-privacy trade-off for differential privacy.
Finally, we perform an ablation study to tease apart which components of
MixMatch are most important for its success.},
added-at = {2019-05-16T10:04:20.000+0200},
author = {Berthelot, David and Carlini, Nicholas and Goodfellow, Ian and Papernot, Nicolas and Oliver, Avital and Raffel, Colin},
biburl = {https://www.bibsonomy.org/bibtex/2a863243733b3b37e4df4588a7fff750c/straybird321},
description = {[1905.02249] MixMatch: A Holistic Approach to Semi-Supervised Learning},
interhash = {3b56ac18ec1dac2793c8871687a4b289},
intrahash = {a863243733b3b37e4df4588a7fff750c},
keywords = {semi-supervised-learning},
note = {cite arxiv:1905.02249},
timestamp = {2019-05-16T10:04:20.000+0200},
title = {MixMatch: A Holistic Approach to Semi-Supervised Learning},
url = {http://arxiv.org/abs/1905.02249},
year = 2019
}