Fair Generative Modeling via Weak Supervision

By Aditya Grover, Kristy Choi, Rui Shu, Stefano Ermon. 2019

171 citations recorded by the source corpus.

Real-world datasets are often biased with respect to key demographic factors such as race and gender. Due to the latent nature of the underlying factors, detecting and mitigating bias is especially challenging for unsupervised machine learning. We present a weakly supervised algorithm for overcoming dataset bias for deep generative models. Our approach requires access to an additional small, unlabeled reference dataset as the supervision signal, thus sidestepping the need for explicit labels on the underlying bias factors. Using this supplementary dataset, we detect the bias in existing datasets via a density ratio technique and learn generative models which efficiently achieve the twin goals of: 1) data efficiency by using training examples from both biased and reference datasets for learning; and 2) data generation close in distribution to the reference dataset at test time. Empirically, we demonstrate the efficacy of our approach which reduces bias w.r.t. latent factors by an average of up to 34.6% over baselines for comparable image generation using generative adversarial networks.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Generative Models, AI Ethics, Fairness & Society

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

In Computer Vision and Pattern Recognition

In Generative Models

In AI Ethics, Fairness & Society

Citations in the corpus

D2C: Diffusion-Decoding Models for Few-Shot Conditional Generation (2021, neutral): “This is in contrast to previous methods that try to learn a fair generative model via a carefully designed representative dataset [20].”
Understanding and Evaluating Racial Biases in Image Captioning (2021, background): “Proposed solutions to dataset bias include new data collection approaches [38, 35], manual data cleanup [74, 75], synthetic data generation [58, 20, 60] – or, in extreme cases, even withdrawing the dataset after insurmountable biases have been identifed [13].”
Pretrained Transformers as Universal Computation Engines (2021, background): “Mitigating these biases is an active area of research (Grover et al. 2019; Choi et al. 2020).”
A Systematic Study of Bias Amplification (2022, methodology): “Prior work has measured bias amplification using generative adversarial networks [8, 22], by considering binary classifications without attributes [28], and by measuring correlations in model predictions [23, 47].”
Data collection and quality challenges in deep learning: a data-centric AI perspective (2021, methodology): “Repair data [62,86,137] Generate data [47,171] Acquire data [23,44,155] In-processing”
Advances, challenges and opportunities in creating data for trustworthy AI (2022, background): “Synthetic data can also exacerbate data disparity when digitally disadvantaged groups are not considered in simulator design 39,40 .”
FairBatch: Batch Selection for Model Fairness (2020, methodology): “…be categorized into (1) pre-processing techniques (Kamiran and Calders, 2011; Zemel et al., 2013; Feldman et al., 2015; du Pin Calmon et al., 2017; Choi et al., 2020; Jiang and Nachum, 2020), which debias or reweight data, (2) in-processing techniques (Kamishima et al., 2012; Zafar et al.,…”
Dialect Diversity in Text Summarization on Twitter (2020, background): “[17] effectively employ small reference image datasets to obtain unbiased image generative models.”
Inclusive GAN: Improving Data and Minority Coverage in Generative Models (2020, background): “[45,41,16], motivated from benefiting a downstream classifier, mainly aim for fair generation conditioned on attribute inputs, in terms of yielding allocative decisions and/or removing the correlation between generation and attribute conditions.”
LARGE: Latent-Based Regression through GAN Semantics (2021, background): “Others enriched a real dataset to improve performance in the low-data domain [4, 16, 59, 73] or on biased or unbalanced data [18,22,33,39,44].”
Fair Attribute Classification through Latent Space De-biasing (2020, methodology): “While [12, 35, 36] require training a new GAN for each bias they want to correct, our method uses a single GAN trained on a biased dataset to augment all attributes.”
Towards Fairness in Visual Recognition: Effective Strategies for Bias Mitigation (2019, background): “Classical literature on mitigating bias generally operates on simpler (often linear) models [11, 50, 28], which are easier to understand and control; only recently have researchers begun looking at mitigating bias in end-to-end trained deep learning models [16, 2, 40, 18, 48, 25, 30, 36, 47, 17].”