arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2204.09583cs.LG
收藏

通过在独立划分上重训练分类器提升组鲁棒性

Improved Group Robustness via Classifier Retraining on Independent Splits

Northeastern University(东北大学)

Thien Hang Nguyen, Hongyang R. Zhang, Huy Le Nguyen

AI总结:

本文提出一种基于训练数据独立划分重训练分类器的简单方法,仅依赖单个超参数即可有效提升深度神经网络的最差组性能,且在图像和文本分类基准上优于 group DRO 和 JTT 等基线。

摘要:

通过最小化平均风险训练的深度神经网络能够取得出色的平均性能。然而,如果某个子群在整体数据总体中代表性不足,其性能可能会下降。组分布鲁棒优化(Sagawa et al., 2020a),简称 group DRO,是学习具有强最差组性能模型的广泛使用的基线。我们注意到,该方法在训练时需要为每个样本提供组标签,并且可能会过拟合小群体,需要强正则化。在训练时组标签数量有限的情况下,Just Train Twice(Liu et al., 2021),简称 JTT,是一种两阶段方法,首先为每个未标记样本推断伪组标签,然后基于推断的组标签应用 group DRO。该推断过程也对过拟合敏感,有时涉及额外的超参数。本文基于在训练数据的独立划分上重训练分类器的思想,设计了一种简单的方法。我们发现,在微调步骤中使用一种新颖的样本划分程序可以实现鲁棒的最差组性能。在基准图像和文本分类任务上进行评估时,无论是在训练期间提供组标签还是仅在验证集中给出组标签,我们的方法在与 group DRO、JTT 及其他强基线相比时均表现良好。重要的是,我们的方法仅依赖于单个超参数,该参数调整用于训练特征提取器与训练分类层的标签比例。我们通过最差组损失的泛化误差界分析论证了该划分方案的合理性。

英文摘要:

Deep neural networks trained by minimizing the average risk can achieve strong average performance. Still, their performance for a subgroup may degrade if the subgroup is underrepresented in the overall data population. Group distributionally robust optimization (Sagawa et al., 2020a), or group DRO in short, is a widely used baseline for learning models with strong worst-group performance. We note that this method requires group labels for every example at training time and can overfit to small groups, requiring strong regularization. Given a limited amount of group labels at training time, Just Train Twice (Liu et al., 2021), or JTT in short, is a two-stage method that infers a pseudo group label for every unlabeled example first, then applies group DRO based on the inferred group labels. The inference process is also sensitive to overfitting, sometimes involving additional hyperparameters. This paper designs a simple method based on the idea of classifier retraining on independent splits of the training data. We find that using a novel sample-splitting procedure achieves robust worst-group performance in the fine-tuning step. When evaluated on benchmark image and text classification tasks, our approach consistently performs favorably to group DRO, JTT, and other strong baselines when either group labels are available during training or are only given in validation sets. Importantly, our method only relies on a single hyperparameter, which adjusts the fraction of labels used for training feature extractors vs. training classification layers. We justify the rationale of our splitting scheme with a generalization-bound analysis of the worst-group loss.

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。

↑