arXiv research

This paper studies the problem of learning with augmented classes (LAC), where augmented classes unobserved in the training data might emerge in the testing phase. Previous studies generally attempt to discover augmented classes by exploiting geometric properties, achieving inspiring empirical performance yet lacking t…

Graph data augmentation improves GNN performance in node classification.

A new data augmentation method selects mixed classes based on class distances for better performance.

This paper evaluates methods to improve classification on imbalanced datasets.

IB-GAN improves multivariate time series classification under imbalance.

The study explores Legendrian fillings and augmentations, providing methods to compute induced augmentations.

Improves tabular data augmentation for contrastive learning.

Synthetic data augmentation can improve imbalanced classification metrics.

Let L be a Legendrian knot in R^3 with the standard contact structure. In [10], a map was constructed from equivalence classes of Morse complex sequences for L, which are combinatorial objects motivated by generating families, to homotopy classes of augmentations of the Legendrian contact homology algebra of L. Moreove…

New theory explains contrastive learning via overlapping augmented views.

Paper proposes an unbiased risk estimator for PLLAC, handling unseen classes.

Effective training of neural networks requires much data. In the low-data regime, parameters are underdetermined, and learnt networks generalise poorly. Data Augmentation alleviates this by using existing data more effectively. However standard data augmentation produces only limited plausible alternative data. Given t…

We propose regularization strategies for learning discriminative models that are robust to in-class variations of the input data. We use the Wasserstein-2 geometry to capture semantically meaningful neighborhoods in the space of images, and define a corresponding input-dependent additive noise data augmentation model. …

Proposes a new framework for learning image augmentations to improve classification performance.

In this paper, we propose a novel implicit semantic data augmentation (ISDA) approach to complement traditional augmentation techniques like flipping, translation or rotation. Our work is motivated by the intriguing property that deep networks are surprisingly good at linearizing features, such that certain directions …

We study the geometry of fully augmented link complements in $S^3$ by looking at their link diagrams. We extend the method introduced by Thistlethwaite and Tsvietkova to fully augmented links and define a system of algebraic equations in terms of parameters coming from edges and crossings of the link diagrams. Combinin…

Computational analysis on physiological signals would provide immense impact for enabling automated clinical analytics. However, the class imbalance issue where negative or minority class instances are rare in number impairs the robustness of the practical solution. The key idea of our approach is intelligent augmentat…

We start by describing how ideal triangulations on a surface degenerate under pinching of a multicurve. We use this process to construct a homomorphism from the Ptolemy groupoid of a surface to that of a pinched surface which is natural with respect to the action of the mapping class group. We then apply this construct…

Deep multi-task learning attracts much attention in recent years as it achieves good performance in many applications. Feature learning is important to deep multi-task learning for sharing common information among tasks. In this paper, we propose a Hierarchical Graph Neural Network (HGNN) to learn augmented features fo…

Compute group cohomology for mapping class group with non-symplectic coefficients.

Self-augmentation improves deep networks for few-shot learning with minimal training data.

We study the fibration of augmented link complements. Given the diagram of an augmented link we associate a spanning surface and a graph. We then show that this surface is a fiber for the link complement if and only if the associated graph is a tree. We further show that fibration is preserved under Dehn filling on cer…

This paper tackles few-shot classification by improving GAN-based data augmentation.

This paper proposes an autoencoder (AE) that is used for improving the performance of once-class classifiers for the purpose of detecting anomalies. Traditional one-class classifiers (OCCs) perform poorly under certain conditions such as high-dimensionality and sparsity. Also, the size of the training set plays an impo…

Regularization and data augmentation can be class-dependent, leading to poor performance on some classes.

We propose a new scalable multi-class Gaussian process classification approach building on a novel modified softmax likelihood function. The new likelihood has two benefits: it leads to well-calibrated uncertainty estimates and allows for an efficient latent variable augmentation. The augmented model has the advantage …

To a Legendrian knot, one can associate an $\mathcal{A}_{\infty}$ category, the augmentation category. An exact Lagrangian cobordism between two Legendrian knots gives a functor of the augmentation categories of the two knots. We study the functor and establish a long exact sequence relating the corresponding cohomolog…

Deep neural networks suffer from over-fitting and catastrophic forgetting when trained with small data. One natural remedy for this problem is data augmentation, which has been recently shown to be effective. However, previous works either assume that intra-class variances can always be generalized to new classes, or e…

Model patching closes subgroup performance gaps in skin cancer classification.

Synthetic augmentation helps but not always in imbalanced learning.

Data augmentation, a technique in which a training set is expanded with class-preserving transformations, is ubiquitous in modern machine learning pipelines. In this paper, we seek to establish a theoretical framework for understanding data augmentation. We approach this from two directions: First, we provide a general…

Data augmentation can achieve the same statistical benefits as full augmentation up to an approximation error.

Data augmentation (DA) is fundamental against overfitting in large convolutional neural networks, especially with a limited training dataset. In images, DA is usually based on heuristic transformations, like geometric or color transformations. Instead of using predefined transformations, our work learns data augmentati…

Herein, we present a system for hyperspectral image segmentation that utilizes multiple class--based denoising autoencoders which are efficiently trained. Moreover, we present a novel hyperspectral data augmentation method for labelled HSI data using linear mixtures of pixels from each class, which helps the system wit…

A new method solves complex constrained minimax problems.

Study challenges the necessity of data augmentation for improving predictions on imbalanced text datasets.

This paper uses GANs to generate synthetic Bitcoin address data.

i-Mix improves contrastive learning across domains without domain-specific augmentations.

We improve the recently-proposed "MixMatch" semi-supervised learning algorithm by introducing two new techniques: distribution alignment and augmentation anchoring. Distribution alignment encourages the marginal distribution of predictions on unlabeled data to be close to the marginal distribution of ground-truth label…

Proves accuracy guarantees for self-supervised learning with correlated positive pairs.

Transforms curves and surfaces for efficient geometric analysis.

The paper analyzes data augmentation for precision matrix estimation in high dimensions.

SONODEs and ANODEs improve learning of second order dynamics.

In this paper we construct an $\mathcal{A}_\infty$-category associated to a Legendrian submanifold of jet spaces. Objects of the category are augmentations of the Chekanov algebra $\mathcal{A}(Λ)$ and the homology of the morphism spaces forms a new set of invariants of Legendrian submanifolds called the bilinearised Le…

MCRAGE generates synthetic data to balance healthcare datasets.

Deep learning has significant potential for medical imaging. However, since the incident rate of each disease varies widely, the frequency of classes in a medical image dataset is imbalanced, leading to poor accuracy for such infrequent classes. One possible solution is data augmentation of infrequent classes using syn…

We give an explicit basis $\mathcal{B}$ of the quotient of the Kauffman bracket skein algebra $\mathcal{S} (Σ)$ on a surface $Σ$ by the square of an augmentation ideal. As an application, it induces two kinds of finite type invariants of links in a handle body in the sense of Le. Moreover, we construct an embedding of …

YuruGAN generates yuru-chara images using GANs and clustering for small datasets.