Semantic Perturbations with Normalizing Flows for Improved Generalization

2021-08-18

INNF+ Workshop, ICML 2021ICML Workshop on Invertible Neural Networks, Normalizing Flows, and Explicit Likelihood Models

Semantic Perturbations with Normalizing Flows for Improved Generalization

Oğuz Kaan Yüksel·Sebastian Urban Stich·Martin Jaggi·Tatjana Chavdarova

August 2021

PaperCode

Research context

Open in graph

Addresses

Computer vision

Uses

Adversarial training Normalizing flow

Studies

Generalization

Produces

AlgorithmData augmentation

Flow-based semantic augmentation

An algorithm that constructs invertible latent-space perturbations for unsupervised data augmentation in image classification.

Versions

• Proceedings of the IEEE International Conference on Computer Vision 2021 · canonical
• ICML Workshop on Invertible Neural Networks, Normalizing Flows, and Explicit Likelihood Models 2021

Related publications

Adversarial Training with Orthogonal Regularization

2020 · SIU

• Related by shared assertionscomputed
• UsesAdversarial training
• StudiesGeneralization

Generalization Bounds for Autoregressive Processes and In-Context Learning

2025 · PriGM@EurIPS

• Related by shared assertionscomputed
• StudiesGeneralization

Re Can gradient clipping mitigate label noise?

2021 · ReScience C

• Related by shared assertionscomputed
• StudiesGeneralization