Manifold Mixup: Better Representations by Interpolating Hidden States

By Vikas Verma, Alex Lamb, Christopher Beckham, Amir Najafi, Ioannis Mitliagkas, David Lopez-Paz, Yoshua Bengio. 2018

1,613 citations recorded by the source corpus.

Subfields: Representation & Self-Supervised Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Representation & Self-Supervised Learning

Citations in the corpus

CutMix: Regularization Strategy to Train Strong Classifiers With Localizable Features (2019, methodology): “Manifold Mixup [41] applies Mixup operation on the randomly chosen internal feature map.”
A U-Net Based Discriminator for Generative Adversarial Networks (2020, background): “[43] extends [49] by performing interpolation not only in the input layer but also in the intermediate layers.”
A Survey of Data Augmentation Approaches for NLP (2021, background): “ing work has explored interpolating inner components (Verma et al., 2019; Faramarzi et al., 2020), more general mixing schemes (Guo, 2020), and adding adversaries (Beckham et al.”
Continual Learning by Modeling Intra-Class Variation (2022, methodology): “, 2017), and representation augmentation methods such as Manifold Mixup (Verma et al., 2019).”
Forward Compatible Few-Shot Class-Incremental Learning (2022, methodology): “Motivated by the intuition that interpolations between two different clusters are often regions of low-confidence predictions [44], we seek to fuse two instances by manifold mixup [44] and treat the fused instance as a virtual new class.”
REMIND Your Neural Network to Prevent Catastrophic Forgetting (2019, methodology): “IV-D. In our experiments, we use the PQ implementation in the Faiss library [35]. A. Augmentation During Replay To augment data during replay, REMIND uses random crops and a variant of manifold mixup [36] for image classiﬁcation on the quantized tensors directly. The tensors are randomly cropped and”
Interpolation Consistency Training for Semi-Supervised Learning (2019, methodology): “To this end, we compare our method against two state-of-the-art supervised learning algorithms (Verma et al., 2018; Zhang et al., 2018), denoted as Supervised(Mixup) and Supervised(Manifold Mixup), respectively in Tables 1 and 2.”
MixText: Linguistically-Informed Interpolation of Hidden Space for Semi-Supervised Text Classification (2020, methodology): “Instead of performing mixup at random input layers like Verma et al. (2019a), choosing which layer of the hidden representations to mixup is an interesting question to investigate.”
Learning Placeholders for Open-Set Recognition (2021, methodology): “Taking the above two characters into consideration, we mimic novel patterns with manifold mixup [31].”
Delving into Deep Imbalanced Regression (2021, positive): “• Manifold-Mixup (M-M IXUP ) (Verma et al., 2019): M-M IXUP extends the idea of M IXUP from input space to the hidden representation space, where the linear interpolations are performed in (multiple) deep hidden layers.”
MedViT: A Robust Vision Transformer for Generalized Medical Image Classification (2023, background): “It is worth pointing out that the decision boundary is often sharp, and there is a significant portion of the hidden representation space that is associated with high-confidence predictions [18, 19].”
A Comprehensive Survey of Image Augmentation Techniques for Deep Learning (2022, methodology): “Simultaneously, we believe that most of the techniques in image augmentation can be extended to feature augmentation, such as Manifold Mixup [114] from Mixup [28] and occluded feature [115].”
BBN: Bilateral-Branch Network With Cumulative Learning for Long-Tailed Visual Recognition (2019, methodology): “Note that, we also conduct experiments with a series of mixup algorithms [33, 27] for comparisons.”
Tackling Catastrophic Forgetting and Background Shift in Continual Semantic Segmentation (2021, background): “Image mixing is popular for classification [93], [94], [95], [96], [97], [98] yet, to the best of our knowledge, sees limited use for semantic segmentation [99], [100], [101], [102], [103], and has never been considered to design memory-efficient rehearsal learning systems.”
On Adversarial Mixup Resynthesis (2019, methodology): “These mixing functions could consist of continuous interpolations between latent vectors such as in Verma et al. (2018), genetically-inspired recombination such as crossover, or even a deep neural network which learns the mixing operation.”
MixCo: Mix-up Contrastive Learning for Visual Representation (2020, methodology): “In this paper, inspired by the success of mix-up training [34] in recent studies [29, 30, 32], we propose Mix-up Contrast (MixCo) as an extension of contrastive learning approach.”