Visual Data Diagnosis and Debiasing with Concept Graphs
This work addresses dataset biases that affect the reliability of deep learning models in visual tasks, representing an incremental advance in debiasing techniques.
The paper tackles the problem of concept co-occurrence biases in visual datasets, which cause unreliable predictions in deep learning models, and presents ConBias, a framework that diagnoses and mitigates these biases using concept graphs and a clique-based balancing strategy, resulting in improved generalization performance across multiple datasets compared to state-of-the-art methods.
The widespread success of deep learning models today is owed to the curation of extensive datasets significant in size and complexity. However, such models frequently pick up inherent biases in the data during the training process, leading to unreliable predictions. Diagnosing and debiasing datasets is thus a necessity to ensure reliable model performance. In this paper, we present ConBias, a novel framework for diagnosing and mitigating Concept co-occurrence Biases in visual datasets. ConBias represents visual datasets as knowledge graphs of concepts, enabling meticulous analysis of spurious concept co-occurrences to uncover concept imbalances across the whole dataset. Moreover, we show that by employing a novel clique-based concept balancing strategy, we can mitigate these imbalances, leading to enhanced performance on downstream tasks. Extensive experiments show that data augmentation based on a balanced concept distribution augmented by Conbias improves generalization performance across multiple datasets compared to state-of-the-art methods.