持续学习中的超参数选择
Hyperparameter Selection in Continual Learning
AI总结:
本文比较多种现实可行的持续学习超参数优化框架,发现它们在常用基准上均无稳定优势,建议采用现实框架并依据计算效率等非性能因素选择。
摘要:
在持续学习(CL)中——学习器在数据流上进行训练——标准超参数优化(HPO)无法应用,因为学习器无法同时访问所有数据。这促使了面向持续学习的 HPO 框架的发展。在 CL 中调整超参数最流行的方法是使用不同的超参数设置在整个数据流上反复训练。然而,这种训练结束时 HPO 在实践中不可用,因为学习器只能看到数据流一次。因此,存在一个悬而未决的问题:在现实中,从业者应当针对 CL 问题使用哪种 HPO 框架?本文通过比较若干现实可行的 HPO 框架来研究这一问题。我们发现,在所考察的 HPO 框架中,包括训练结束时 HPO 在内,没有任何框架能在流行的 CL 基准测试上始终优于其他框架。因此,我们得出两点结论:a)为了能够区分不同 HPO 框架,需要超越当前最常用的 CL 基准测试;b)在所考察的流行 CL 基准测试上,CL 从业者应使用现实可行的 HPO 框架,并且可以根据性能之外的因素来选择,例如计算效率。
英文摘要:
In continual learning (CL) -- where a learner trains on a stream of data -- standard hyperparameter optimisation (HPO) cannot be applied, as a learner does not have access to all of the data at the same time. This has prompted the development of CL-specific HPO frameworks. The most popular way to tune hyperparameters in CL is to repeatedly train over the whole data stream with different hyperparameter settings. However, this end-of-training HPO is unusable in practice since a learner can only see the stream once. Hence, there is an open question: what HPO framework should a practitioner use for a CL problem in reality? This paper looks at this question by comparing several realistic HPO frameworks. We find that none of the HPO frameworks considered, including end-of-training HPO, perform consistently better than the rest on popular CL benchmarks. We therefore arrive at a twofold conclusion: a) to be able to discriminate between HPO frameworks there is a need to move beyond the current most commonly used CL benchmarks, and b) on the popular CL benchmarks examined, a CL practitioner should use a realistic HPO framework and can select it based on factors separate from performance, for example compute efficiency.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。