TaskMix: Data Augmentation for Meta-Learning of Spoken Intent Understanding

TL;DR

This paper introduces TaskMix, a data augmentation technique for meta-learning in spoken intent understanding, which synthesizes new tasks to improve transfer learning and reduce overfitting, especially with low task diversity.

Contribution

TaskMix is a simple yet effective method that interpolates existing tasks to enhance meta-learning for spoken intent classification, outperforming baselines in low-diversity scenarios.

Findings

TaskMix improves meta-learning performance on intent classification datasets.

It alleviates overfitting when task diversity is low.

It maintains performance even with high task diversity.

Abstract

Meta-Learning has emerged as a research direction to better transfer knowledge from related tasks to unseen but related tasks. However, Meta-Learning requires many training tasks to learn representations that transfer well to unseen tasks; otherwise, it leads to overfitting, and the performance degenerates to worse than Multi-task Learning. We show that a state-of-the-art data augmentation method worsens this problem of overfitting when the task diversity is low. We propose a simple method, TaskMix, which synthesizes new tasks by linearly interpolating existing tasks. We compare TaskMix against many baselines on an in-house multilingual intent classification dataset of N-Best ASR hypotheses derived from real-life human-machine telephony utterances and two datasets derived from MTOP. We show that TaskMix outperforms baselines, alleviates overfitting when task diversity is low, and does not degrade performance even when it is high.