Addressing Model Vulnerability to Distributional Shifts Over Image Transformation Sets

Volpi, Riccardo; Murino, Vittorio

doi:10.1109/iccv.2019.00807

Cited by 66 publications

(61 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…flipping and rotation. However, conventional data augmentation methods only deal with simple geometric changes within the same dataset (Volpi and Murino 2019). When the domain gap is large such as those illustrated in Figure 4 containing image style variations, learning-based augmentation strategies are required.…”

Section: Related Workmentioning

confidence: 99%

Deep Domain-Adversarial Image Generation for Domain Generalisation

Zhou

Yang

Hospedales

et al. 2020

AAAI

235

156

View full text Add to dashboard Cite

Machine learning models typically suffer from the domain shift problem when trained on a source dataset and evaluated on a target dataset of different distribution. To overcome this problem, domain generalisation (DG) methods aim to leverage data from multiple source domains so that a trained model can generalise to unseen domains. In this paper, we propose a novel DG approach based on Deep Domain-Adversarial Image Generation (DDAIG). Specifically, DDAIG consists of three components, namely a label classifier, a domain classifier and a domain transformation network (DoTNet). The goal for DoTNet is to map the source training data to unseen domains. This is achieved by having a learning objective formulated to ensure that the generated data can be correctly classified by the label classifier while fooling the domain classifier. By augmenting the source training data with the generated unseen domain data, we can make the label classifier more robust to unknown domain changes. Extensive experiments on four DG datasets demonstrate the effectiveness of our approach.

show abstract

Section: Related Workmentioning

confidence: 99%

Deep Domain-Adversarial Image Generation for Domain Generalisation

Zhou

Yang

Hospedales

et al. 2020

AAAI

235

156

View full text Add to dashboard Cite

show abstract

“…5 in order to maximize validation performance [139]. Since augmentation operations are often non-differentiable, this requires reinforcement learning [139], discrete gradient-estimators [140], or evolutionary [141] methods. Recent attempts use meta-gradient to learn mixing proportions in mixup-based augmentation [142].…”

Section: Embedding Functions (Metric Learning)mentioning

confidence: 99%

“…EAs are relatively more commonly applied in RL applications [24], [168] (where models are typically smaller, and inner optimizations are long and non-differentiable). However they have also been applied to learn learning rules [194], optimizers [195], architectures [26], [126] and data augmentation strategies [141] in supervised learning. They are also particularly important in learning human interpretable symbolic meta-representations [119].…”

Section: Evolutionmentioning

confidence: 99%

Meta-Learning in Neural Networks: A Survey

Hospedales¹,

Antoniou²,

Micaelli³

et al. 2021

IEEE Trans. Pattern Anal. Mach. Intell.

661

371

View full text Add to dashboard Cite

The field of meta-learning, or learning-to-learn, has seen a dramatic rise in interest in recent years. Contrary to conventional approaches to AI where tasks are solved from scratch using a fixed learning algorithm, meta-learning aims to improve the learning algorithm itself, given the experience of multiple learning episodes. This paradigm provides an opportunity to tackle many conventional challenges of deep learning, including data and computation bottlenecks, as well as generalization. This survey describes the contemporary meta-learning landscape. We first discuss definitions of meta-learning and position it with respect to related fields, such as transfer learning and hyperparameter optimization. We then propose a new taxonomy that provides a more comprehensive breakdown of the space of meta-learning methods today. We survey promising applications and successes of meta-learning such as few-shot learning and reinforcement learning. Finally, we discuss outstanding challenges and promising areas for future research.

show abstract

“…Existing works on single domain generalization [20,42,52,53,60] try to improve the generalization capability through adversarial domain augmentation (ADA), which synthesizes new training images in an adversarial way to mimic virtual challenging domains. The model therefore learns the domain-invariant features to improve its generalization performance.…”

Section: Introductionmentioning

confidence: 99%

“…To this end, we propose a novel adaptive form of normal-ization named as adaptive standardization and rescaling normalization (ASR-Norm), in which the standardization and rescaling statistics are both learned to be adaptive to each individual input sample. When being used with ADA [20,52,53], ASR-Norm can learn the normalization statistics by approximately optimizing a robust objective, making the statistics be adaptive to the data coming from different domains, and hence helping the model to generalize better across domains than traditional normalization approaches. We also show that ASR-Norm can be viewed as a generic form of the traditional normalization approaches including BN, IN, layer normalization (LN) [1], group normalization (GN) [55], and switchable normalization (SN) [32].…”

Section: Introductionmentioning

confidence: 99%

Adversarially Adaptive Normalization for Single Domain Generalization

Wang¹,

Yang²,

Gong³

et al. 2021

Preprint

View full text Add to dashboard Cite

Single domain generalization aims to learn a model that performs well on many unseen domains with only one domain data for training. Existing works focus on studying the adversarial domain augmentation (ADA) to improve the model's generalization capability. The impact on domain generalization of the statistics of normalization layers is still underinvestigated. In this paper, we propose a generic normalization approach, adaptive standardization and rescaling normalization (ASR-Norm), to complement the missing part in previous works. ASR-Norm learns both the standardization and rescaling statistics via neural networks. This new form of normalization can be viewed as a generic form of the traditional normalizations. When trained with ADA, the statistics in ASR-Norm are learned to be adaptive to the data coming from different domains, and hence improves the model generalization performance across domains, especially on the target domain with large discrepancy from the source domain. The experimental results show that ASR-Norm can bring consistent improvement to the state-of-the-art ADA approaches by 1.6%, 2.7%, and 6.3% averagely on the Digits, CIFAR-10-C, and PACS benchmarks, respectively. As a generic tool, the improvement introduced by ASR-Norm is agnostic to the choice of ADA methods. * The main work was done during an internship at Google Research. Figure 1: Illustration of single domain generalization with the PACS [30] benchmark. The dataset contains 4 domains: art paint, cartoon, sketch, and photo domains, which share the same categories that include dog, elephant, giraffe, guitar, house, horse, and person. Single domain generalization aims at training a model on one source domain data (art paint domain in the shown case), while generalizing well to other domains with very different visual presentations.

show abstract

Addressing Model Vulnerability to Distributional Shifts Over Image Transformation Sets

Cited by 66 publications

References 25 publications

Deep Domain-Adversarial Image Generation for Domain Generalisation

Deep Domain-Adversarial Image Generation for Domain Generalisation

Meta-Learning in Neural Networks: A Survey

Adversarially Adaptive Normalization for Single Domain Generalization

Contact Info

Product

Resources

About