Papers of the day   All papers

MixMatch: A Holistic Approach to Semi-Supervised Learning

Comments

May 15 2019 David Berthelot

MixMatch (https://arxiv.org/abs/1905.02249) code is released https://github.com/google-research/mixmatch (Python3, TensorFlow 1.1x). Let me know how it works for you and also let me know if you port it to other frameworks.
3 replies, 460 likes


May 08 2019 David Berthelot

New paper: MixMatch: A Holistic Approach to Semi-Supervised Learning http://arxiv.org/abs/1905.02249 Reduces error rate by up to 4x on CIFAR10. With Nicholas Carlini @goodfellow_ian @NicolasPapernot @avitaloliver @colinraffel https://t.co/TawK9E3I5e
6 replies, 413 likes


May 08 2019 Nicolas Papernot

In addition to reducing the need for labeled data, developments in semi-supervised learning have made it easier to learn with differential privacy using PATE: @D_Berthelot_ML's MixMatch approach to semi-supervised learning significantly improves the state-of-the-art
2 replies, 136 likes


May 18 2019 Quoc Le

Links to the mentioned papers. MixMatch: https://arxiv.org/abs/1905.02249 Unsupervised Data Augmentation: https://arxiv.org/abs/1904.12848
1 replies, 82 likes


May 16 2019 Avital Oliver

Here's the code behind MixMatch, @D_Berthelot_ML's recent semi-supervised learning method. Big improvements to common benchmarks. Basic idea: 1/ target labels for unlabeled data: avg 2 predictions of model (resample noise) 2/ sharpen target labels 3/ mixup labeled+unlabeled
2 replies, 82 likes


May 15 2019 Nicolas Papernot

Code for MixMatch was just released on GitHub by @D_Berthelot_ML Includes the setup needed to train a PATE student with MixMatch
0 replies, 52 likes


May 18 2019 Olivier Grisel

Mixmatch gives really impressive results on semi supervised benchmark classification tasks with few labels. Looking forward to seeing it be adapted to more useful tasks and data such as object detection/segmentation in medical images for instance.
1 replies, 48 likes


May 16 2019 DataScienceNigeria

You need to check out MixMatch, a new algorithm for training models on a small amount of labeled data & a large amount of unlabeled data that is far more accurate than other methods Great work by @D_Berthelot_ML etal Code:https://github.com/google-research/mixmatch Paper: https://arxiv.org/pdf/1905.02249.pdf https://t.co/Mk0aOsE9SR
0 replies, 36 likes


May 16 2019 mat kelcey

Really enjoyed the MixMatch paper by @D_Berthelot_ML et al. Introduced lots of ideas new to me. I've had a lot of success with trivial pseudo labelling so looking forward to trying some of these concepts! (Though my y isn't just a simple distribution) https://arxiv.org/abs/1905.02249
0 replies, 18 likes


May 23 2019 Motoki Wu 🍤

Artificial data as a smarter version of dropout. In our models, we don't even bother with tuning dropout since tuning artificial data has a much stronger effect. https://arxiv.org/abs/1905.02249
1 replies, 17 likes


May 20 2019 Miles Brundage

Late to the party on this one, but this is a good blog post: https://towardsdatascience.com/the-quiet-semi-supervised-revolution-edec1e9ad8c
0 replies, 16 likes


Jul 19 2019 Vaishaal

This is an INSANE semi supervised result. They get 89 percent accuracy on CIFAR-10 (training set size 50k) with only 250 labeled points, and 95 percent with 4000 labeled points. https://arxiv.org/abs/1905.02249
0 replies, 15 likes


May 17 2019 Piotr Czapla

MixMatch - as easy to understand as MixUp and super powerful way to train models using just a few examples. The code was released yesterday. https://github.com/google-research/mixmatch
0 replies, 11 likes


May 15 2019 David Berthelot

Code on GitHub https://twitter.com/D_Berthelot_ML/status/1128771013074751489
0 replies, 6 likes


May 09 2019 arxiv

MixMatch: A Holistic Approach to Semi-Supervised Learning. http://arxiv.org/abs/1905.02249 https://t.co/retmZX1WD7
0 replies, 5 likes


May 16 2019 Daisuke Okanohara

For semi-supervised learning, MixMatch applies multiple data augmentation and takes an average of estimations, and shapen the distribution by lowering the temperature, then applies MixUp. It significantly improves SSL with few labeled data. https://arxiv.org/abs/1905.02249
0 replies, 4 likes


May 08 2019 Brundage Bot

MixMatch: A Holistic Approach to Semi-Supervised Learning. David Berthelot, Nicholas Carlini, Ian Goodfellow, Nicolas Papernot, Avital Oliver, and Colin Raffel http://arxiv.org/abs/1905.02249
1 replies, 3 likes


Oct 27 2019 arXiv CS-CV

MixMatch: A Holistic Approach to Semi-Supervised Learning http://arxiv.org/abs/1905.02249
0 replies, 2 likes


Oct 26 2019 arXiv CS-CV

MixMatch: A Holistic Approach to Semi-Supervised Learning http://arxiv.org/abs/1905.02249
0 replies, 2 likes


Oct 26 2019 arXiv CS-CV

MixMatch: A Holistic Approach to Semi-Supervised Learning http://arxiv.org/abs/1905.02249
0 replies, 1 likes


Oct 25 2019 arXiv CS-CV

MixMatch: A Holistic Approach to Semi-Supervised Learning http://arxiv.org/abs/1905.02249
0 replies, 1 likes


May 08 2019 那須音トウ:🍆🕸

MixMatch: A Holistic Approach to Semi-Supervised Learning https://arxiv.org/abs/1905.02249
1 replies, 1 likes


Content