Generalizing Under Distribution Shifts and Data Scarcity via ...

Generalizing Under Distribution Shifts and Data Scarcity via ... - KEEP

This dissertation presents novel solutions for improving the generalization capabilities of deep learning based computer vision models.

The dissertation also presents algorithms for enabling in-the-wild generalization without needing access to any samples from the target domain. Other causes of ...

Estimating Generalization under Distribution Shifts via Domain ...

In many applications, machine learning models are deployed on data whose distribution is different from that of the train- ing data. For instance, self-driving ...

A Survey on Evaluation of Out-of-Distribution Generalization - arXiv

It is worth investigating how data generation through SOTA generative models can help in creating OOD datasets. Moreover, despite the richness ...

Delving Deep into the Generalization of Vision Transformers under ...

define a taxonomy on data distribution shifts according to the modified semantic concepts in images. ... to unseen domains via adversarial data augmentation.

An Empirical Study on Distribution Shift Robustness From the ...

our paper evaluates on the WILDS benchmark in addition to the domain generalization data. Similar to our GroupMix, [16] proposes to use a selective mix-up ...

(PDF) Robust Generalization despite Distribution Shift via Minimum ...

Training models that perform well under distribution shifts is a central challenge in machine learning. In this paper, we introduce a ...

DomainVerse: A Benchmark Towards Real-World Distribution Shifts ...

To tackle the aforementioned challenge, domain generalization (DG) (Xu et al., 2021; Hou et al., 2023) is proposed to train a source model on ...

Test-Time Training with Self-Supervision for Generalization under ...

Supervised learning remains notoriously weak at generaliza- tion under distribution shifts. Unless training and test data are drawn from the same distribution, ...

Generative models improve fairness of medical classifiers under ...

We show that diffusion models can automatically learn realistic augmentations from data in a label-efficient manner. We demonstrate that learned ...

Estimating Generalization under Distribution Shifts via Domain ...

Request PDF | Estimating Generalization under Distribution Shifts via Domain-Invariant Representations | When machine learning models are deployed on a test ...

Domain generalization via optimal transport with metric similarity ...

Generalizing knowledge to unseen domains, where data and labels are unavailable, is crucial for machine learning models.

Scarce data driven deep learning of drones via generalized data ...

Generative models are designed to generate data with the same FD as that learned by NNs. In principle, most common generative models, including ...

Near-Optimal Linear Regression under Distribution Shift

Transfer learning is essential when sufficient data comes from the source domain, with scarce la- beled data from the target domain. We develop.

3.6. Generalization — Dive into Deep Learning 1.0.3 documentation

The phenomenon of fitting closer to our training data than to the underlying distribution is called overfitting, and techniques for combatting overfitting are ...

Assaying Out-Of-Distribution Generalization in Transfer Learning

... generalization), especially when data is scarce. While fine-tuning the ... data but seem ill-suited to mimic natural distribution. 9. Page 10. shifts. (4) ...

Model-Agnostic Random Weighting for Out-of-Distribution ...

This challenge has instigated recent research endeavors focusing on out-of-distribution (OOD) generalization. A particularly pervasive and ...

Domain Generalization in Vision: A Survey

However, due to distribution shift between different patients' data, a clas- sifier learned using data from historic patients does not generalize to new ...

Federated Learning under Covariate Shifts with Generalization ...

imbalanced federated settings in terms of data distribution shifts across clients. ... methods aim at minimizing the target risk through data in the source ...

Deep diversity learning for better generalization to unseen domains

However, deep neural networks are susceptible to data distribution shifts occurring between training and use. This apparent flaw prevents the ...