Simple data balancing achieves competitive worst-group-accuracy

27 Oct 2021  ยท  Badr Youbi Idrissi, Martin Arjovsky, Mohammad Pezeshki, David Lopez-Paz ยท

We study the problem of learning classifiers that perform well across (known or unknown) groups of data. After observing that common worst-group-accuracy datasets suffer from substantial imbalances, we set out to compare state-of-the-art methods to simple balancing of classes and groups by either subsampling or reweighting data. Our results show that these data balancing baselines achieve state-of-the-art-accuracy, while being faster to train and requiring no additional hyper-parameters. In addition, we highlight that access to group information is most critical for model selection purposes, and not so much during training. All in all, our findings beg closer examination of benchmarks and methods for research in worst-group-accuracy optimization.

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Out-of-Distribution Generalization UrbanCars SUBG (CoObj) BG Gap -60.2 # 1
CoObj Gap +2.5 # 1
BG+CoObj Gap -62.4 # 1
Out-of-Distribution Generalization UrbanCars SUBG (BG) BG Gap +1.3 # 1
CoObj Gap -36.4 # 1
BG+CoObj Gap -35.8 # 1
Out-of-Distribution Generalization UrbanCars SUBG (BG+CoObj) BG Gap -4.7 # 1
CoObj Gap -0.3 # 1
BG+CoObj Gap -6.3 # 1

Methods


No methods listed for this paper. Add relevant methods here