ترغب بنشر مسار تعليمي؟ اضغط هنا

Epsilon Consistent Mixup: An Adaptive Consistency-Interpolation Tradeoff

69   0   0.0 ( 0 )
 نشر من قبل Vincent Pisztora
 تاريخ النشر 2021
والبحث باللغة English




اسأل ChatGPT حول البحث

In this paper we propose $epsilon$-Consistent Mixup ($epsilon$mu). $epsilon$mu is a data-based structural regularization technique that combines Mixups linear interpolation with consistency regularization in the Mixup direction, by compelling a simple adaptive tradeoff between the two. This learnable combination of consistency and interpolation induces a more flexible structure on the evolution of the response across the feature space and is shown to improve semi-supervised classification accuracy on the SVHN and CIFAR10 benchmark datasets, yielding the largest gains in the most challenging low label-availability scenarios. Empirical studies comparing $epsilon$mu and Mixup are presented and provide insight into the mechanisms behind $epsilon$mus effectiveness. In particular, $epsilon$mu is found to produce more accurate synthetic labels and more confident predictions than Mixup.

قيم البحث

اقرأ أيضاً

MixUp is an effective data augmentation method to regularize deep neural networks via random linear interpolations between pairs of samples and their labels. It plays an important role in model regularization, semi-supervised learning and domain adap tion. However, despite its empirical success, its deficiency of randomly mixing samples has poorly been studied. Since deep networks are capable of memorizing the entire dataset, the corrupted samples generated by vanilla MixUp with a badly chosen interpolation policy will degrade the performance of networks. To overcome the underfitting by corrupted samples, inspired by Meta-learning (learning to learn), we propose a novel technique of learning to mixup in this work, namely, MetaMixUp. Unlike the vanilla MixUp that samples interpolation policy from a predefined distribution, this paper introduces a meta-learning based online optimization approach to dynamically learn the interpolation policy in a data-adaptive way. The validation set performance via meta-learning captures the underfitting issue, which provides more information to refine interpolation policy. Furthermore, we adapt our method for pseudo-label based semisupervised learning (SSL) along with a refined pseudo-labeling strategy. In our experiments, our method achieves better performance than vanilla MixUp and its variants under supervised learning configuration. In particular, extensive experiments show that our MetaMixUp adapted SSL greatly outperforms MixUp and many state-of-the-art methods on CIFAR-10 and SVHN benchmarks under SSL configuration.
We show that minimum-norm interpolation in the Reproducing Kernel Hilbert Space corresponding to the Laplace kernel is not consistent if input dimension is constant. The lower bound holds for any choice of kernel bandwidth, even if selected based on data. The result supports the empirical observation that minimum-norm interpolation (that is, exact fit to training data) in RKHS generalizes well for some high-dimensional datasets, but not for low-dimensional ones.
Recently proposed consistency-based Semi-Supervised Learning (SSL) methods such as the $Pi$-model, temporal ensembling, the mean teacher, or the virtual adversarial training, have advanced the state of the art in several SSL tasks. These methods can typically reach performances that are comparable to their fully supervised counterparts while using only a fraction of labelled examples. Despite these methodological advances, the understanding of these methods is still relatively limited. In this text, we analyse (variations of) the $Pi$-model in settings where analytically tractable results can be obtained. We establish links with Manifold Tangent Classifiers and demonstrate that the quality of the perturbations is key to obtaining reasonable SSL performances. Importantly, we propose a simple extension of the Hidden Manifold Model that naturally incorporates data-augmentation schemes and offers a framework for understanding and experimenting with SSL methods.
The autoencoder model uses an encoder to map data samples to a lower dimensional latent space and then a decoder to map the latent space representations back to the data space. Implicitly, it relies on the encoder to approximate the inverse of the de coder network, so that samples can be mapped to and back from the latent space faithfully. This approximation may lead to sub-optimal latent space representations. In this work, we investigate a decoder-only method that uses gradient flow to encode data samples in the latent space. The gradient flow is defined based on a given decoder and aims to find the optimal latent space representation for any given sample through optimisation, eliminating the need of an approximate inversion through an encoder. Implementing gradient flow through ordinary differential equations (ODE), we leverage the adjoint method to train a given decoder. We further show empirically that the costly integrals in the adjoint method may not be entirely necessary. Additionally, we propose a $2^{nd}$ order ODE variant to the method, which approximates Nesterovs accelerated gradient descent, with faster convergence per iteration. Commonly used ODE solvers can be quite sensitive to the integration step-size depending on the stiffness of the ODE. To overcome the sensitivity for gradient flow encoding, we use an adaptive solver that prioritises minimising loss at each integration step. We assess the proposed method in comparison to the autoencoding model. In our experiments, GFE showed a much higher data-efficiency than the autoencoding model, which can be crucial for data scarce applications.
We consider a problem of multiclass classification, where the training sample $S_n = {(X_i, Y_i)}_{i=1}^n$ is generated from the model $mathbb P(Y = m | X = x) = eta_m(x)$, $1 leq m leq M$, and $eta_1(x), dots, eta_M(x)$ are unknown $alpha$-Holder co ntinuous functions.Given a test point $X$, our goal is to predict its label. A widely used $mathsf k$-nearest-neighbors classifier constructs estimates of $eta_1(X), dots, eta_M(X)$ and uses a plug-in rule for the prediction. However, it requires a proper choice of the smoothing parameter $mathsf k$, which may become tricky in some situations. In our solution, we fix several integers $n_1, dots, n_K$, compute corresponding $n_k$-nearest-neighbor estimates for each $m$ and each $n_k$ and apply an aggregation procedure. We study an algorithm, which constructs a convex combination of these estimates such that the aggregated estimate behaves approximately as well as an oracle choice. We also provide a non-asymptotic analysis of the procedure, prove its adaptation to the unknown smoothness parameter $alpha$ and to the margin and establish rates of convergence under mild assumptions.

الأسئلة المقترحة

التعليقات
جاري جلب التعليقات جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
mircosoft-partner

هل ترغب بارسال اشعارات عن اخر التحديثات في شمرا-اكاديميا