2023 · 19 citations · 25 references
ComputingComputer ScienceData ScienceComputer VisionMachine LearningStatistical LearningBiasBias ControlRepresentation LearningResampling MethodsResampling TechniquesUnderfittingBias MitigationClass ImbalanceClass Imbalance LearningDataset BiasParameter TuningRepresentation Of Underrepresented GroupsSimple Random SamplingSpurious Correlation
Prior work has shown that Visual Recognition datasets frequently underrepresent bias groups <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$B$</tex> (e.g. Female) within class labels <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$Y$</tex> (e.g. Programmers). This dataset bias can lead to models that learn spurious correlations between class labels and bias groups such as age, gender, or race. Most recent methods that address this problem require significant architectural changes or additional loss functions requiring more hyper-parameter tuning. Alternatively, data sampling baselines from the class imbalance literature (e.g. Undersampling, Upweighting), which can often be implemented in a single line of code and often have no hyperparameters, offer a cheaper and more efficient solution. However, these methods suffer from significant shortcomings. For example, Undersampling drops a significant part of the input distribution per epoch while Oversampling repeats samples, causing overfitting. To address these shortcomings, we introduce a new class-conditioned sampling method: Bias Mimicking. The method is based on the observation that if a class <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$c$</tex> bias distribution, i.e. <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$P_{D}(B\vert Y=c)$</tex> is mimicked across every <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$c^{\prime}\neq c$</tex> , then <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$Y$</tex> and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$B$</tex> are statistically independent. Using this notion, BM, through a novel training procedure, ensures that the model is exposed to the entire distribution per epoch without repeating samples. Consequently, Bias Mimicking improves underrepresented groups' accuracy of sampling methods by 3% over four benchmarks while maintaining and sometimes improving performance over nonsampling methods. Code: https://github.com/mqraitem/Bias-Mimicking
25
SMOTE: Synthetic Minority Over-sampling Technique
Nitesh V. Chawla, Kevin W. Bowyer, Lawrence Hall et al. · Journal of Artificial Intelligence Research · 2002 · 31.2K citations · Full text
Momentum Contrast for Unsupervised Visual Representation Learning
Kaiming He, Haoqi Fan, Yuxin Wu et al. · 2020 · 12K citations
Deep Learning Face Attributes in the Wild
Ziwei Liu, Ping Luo, Xiaogang Wang et al. · 2015 · 7.7K citations
The class imbalance problem: A systematic study1
Nathalie Japkowicz, Shaju Stephen · Intelligent Data Analysis · 2002 · 3.3K citations