Paper Image

Tailoring data mixing with similarity

Published on:

2 November 2023

Primary Category:

Machine Learning

Paper Authors:

Quentin Bouniot,

Pavlo Mozharovskyi,

Florence d'Alché-Buc


Key Details

Proposes warping functions to change mixup interpolation distributions

Uses similarity kernels to find warping parameters for each data pair

Can be applied to classification and regression tasks

Improves accuracy and calibration over vanilla mixup

Competitive with state-of-the-art methods while more efficient

AI generated summary

Tailoring data mixing with similarity

This paper proposes a new method to improve data augmentation through mixup. It tailors the interpolation strength when mixing training examples based on their similarity. More similar examples are mixed more strongly. This helps generate useful synthetic data and improves model performance and calibration.

Answers from this paper


No comments yet, be the first to start the conversation...

Sign up to comment on this paper

Sign Up