Learning De-biased Representations with Biased Representations

By Hyojin Bahng, Sanghyuk Chun, Sangdoo Yun, J. Choo, Seong Joon Oh. 2019

339 citations recorded by the source corpus.

Many machine learning algorithms are trained and evaluated by splitting data from a single source into training and test sets. While such focus on in-distribution learning scenarios has led to interesting advancement, it has not been able to tell if models are relying on dataset biases as shortcuts for successful prediction (e.g., using snow cues for recognising snowmobiles), resulting in biased models that fail to generalise when the bias shifts to a different class. The cross-bias generalisation problem has been addressed by de-biasing training data through augmentation or re-sampling, which are often prohibitive due to the data collection cost (e.g., collecting images of a snowmobile on a desert) and the difficulty of quantifying or expressing biases in the first place. In this work, we propose a novel framework to train a de-biased representation by encouraging it to be different from a set of representations that are biased by design. This tactic is feasible in many scenarios where it is much easier to define a set of biased representations than to define and quantify bias. We demonstrate the efficacy of our method across a variety of synthetic and real-world biases; our experiments show that the method discourages models from taking bias shortcuts, resulting in improved generalisation. Source code is available at this https URL.

Subfields: Artificial Intelligence

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Bias in machine learning - what is it good for? (2020, neutral): “A similar effect is reported in [2].”
Deep Stable Learning for Out-Of-Distribution Generalization (2021, methodology): “Hilbert-Schmidt Independence Criterion (HSIC) [18], which requires that the squared Hilbert-Schmidt norm of ΣAB should be zero, can be applied as a criterion to supervise feature decorrelation [3].”
Last Layer Re-Training is Sufficient for Robustness to Spurious Correlations (2022, background): “Prior work has often associated poor robustness to spurious correlations with the quality of representations learned by the model [4, 6, 84] and suggested that the entire model needs to be carefully trained to”