Training Dynamics for Curriculum Learning: A Study on Monolingual and Cross-lingual NLU
This work addresses enhancing model generalization and efficiency in NLU tasks, particularly for cross-lingual and OOD scenarios, but it is incremental as it modifies existing methods.
The paper tackled improving curriculum learning for natural language understanding by using training dynamics as difficulty metrics, resulting in up to 8.5% better performance in zero-shot cross-lingual and out-of-distribution settings and 20% faster training on average.
Curriculum Learning (CL) is a technique of training models via ranking examples in a typically increasing difficulty trend with the aim of accelerating convergence and improving generalisability. Current approaches for Natural Language Understanding (NLU) tasks use CL to improve in-distribution data performance often via heuristic-oriented or task-agnostic difficulties. In this work, instead, we employ CL for NLU by taking advantage of training dynamics as difficulty metrics, i.e., statistics that measure the behavior of the model at hand on specific task-data instances during training and propose modifications of existing CL schedulers based on these statistics. Differently from existing works, we focus on evaluating models on in-distribution (ID), out-of-distribution (OOD) as well as zero-shot (ZS) cross-lingual transfer datasets. We show across several NLU tasks that CL with training dynamics can result in better performance mostly on zero-shot cross-lingual transfer and OOD settings with improvements up by 8.5% in certain cases. Overall, experiments indicate that training dynamics can lead to better performing models with smoother training compared to other difficulty metrics while being 20% faster on average. In addition, through analysis we shed light on the correlations of task-specific versus task-agnostic metrics.