# Benchmarking Algorithmic Bias in Face Recognition: An Experimental Approach Using Synthetic Faces and Human Evaluation

arXiv: https://arxiv.org/abs/2308.05441 · HTML: https://mlsanity.com/paper/2308.05441/

## Key facts

- Cited by 25 papers in the map (data through 14 Oct 2026).
- Builds on 33 of its 57 references that are in the map.
- First posted on arXiv 10 Aug 2023.

## Builds on (most cited first, 30 of 33)

- [Denoising Diffusion Probabilistic Models](https://mlsanity.com/paper/2006.11239/) (2020)
- [A Style-Based Generator Architecture for Generative Adversarial Networks](https://mlsanity.com/paper/1812.04948/) (2018)
- [Generative Adversarial Networks](https://mlsanity.com/paper/2203.00667/) (2022)
- [Neural Discrete Representation Learning](https://mlsanity.com/paper/1711.00937/) (2017)
- [Improved Denoising Diffusion Probabilistic Models](https://mlsanity.com/paper/2102.09672/) (2021)
- [Deep Learning Face Attributes in the Wild](https://mlsanity.com/paper/1411.7766/) (2014)
- [Analyzing and Improving the Image Quality of StyleGAN](https://mlsanity.com/paper/1912.04958/) (2019)
- [FaceNet: A Unified Embedding for Face Recognition and Clustering](https://mlsanity.com/paper/1503.03832/) (2015)
- [Equality of Opportunity in Supervised Learning](https://mlsanity.com/paper/1610.02413/) (2016)
- [Efficient Geometry-aware 3D Generative Adversarial Networks](https://mlsanity.com/paper/2112.07945/) (2021)
- [VGGFace2: A dataset for recognising faces across pose and age](https://mlsanity.com/paper/1710.08092/) (2017)
- [SphereFace: Deep Hypersphere Embedding for Face Recognition](https://mlsanity.com/paper/1704.08063/) (2017)
- [Encoding in Style: a StyleGAN Encoder for Image-to-Image Translation](https://mlsanity.com/paper/2008.00951/) (2020)
- [Interpreting the Latent Space of GANs for Semantic Face Editing](https://mlsanity.com/paper/1907.10786/) (2019)
- [Mitigating Unwanted Biases with Adversarial Learning](https://mlsanity.com/paper/1801.07593/) (2018)
- [BiSeNet: Bilateral Segmentation Network for Real-time Semantic Segmentation](https://mlsanity.com/paper/1808.00897/) (2018)
- [GANSpace: Discovering Interpretable GAN Controls](https://mlsanity.com/paper/2004.02546/) (2020)
- [Preventing Fairness Gerrymandering: Auditing and Learning for Subgroup Fairness](https://mlsanity.com/paper/1711.05144/) (2017)
- [Partial FC: Training 10 Million Identities on a Single Machine](https://mlsanity.com/paper/2010.05222/) (2020)
- [Fair Attribute Classification through Latent Space De-biasing](https://mlsanity.com/paper/2012.01469/) (2020)
- [Fairness in Deep Learning: A Computational Perspective](https://mlsanity.com/paper/1908.08843/) (2019)
- [Killing Two Birds with One Stone:Efficient and Robust Training of Face Recognition CNNs by Partial FC](https://mlsanity.com/paper/2203.15565/) (2022)
- [Real-time Facial Surface Geometry from Monocular Video on Mobile GPUs](https://mlsanity.com/paper/1907.06724/) (2019)
- [Towards causal benchmarking of bias in face analysis algorithms](https://mlsanity.com/paper/2007.06570/) (2020)
- [Leveling Down in Computer Vision: Pareto Inefficiencies in Fair Deep Classifiers](https://mlsanity.com/paper/2203.04913/) (2022)
- [SphereFace2: Binary Classification is All You Need for Deep Face Recognition](https://mlsanity.com/paper/2108.01513/) (2021)
- [Fairness in Biometrics: a figure of merit to assess biometric verification systems](https://mlsanity.com/paper/2011.02395/) (2020)
- [How Does Gender Balance In Training Data Affect Face Recognition Accuracy?](https://mlsanity.com/paper/2002.02934/) (2020)
- [Gendered Differences in Face Recognition Accuracy Explained by Hairstyles, Makeup, and Facial Morphology](https://mlsanity.com/paper/2112.14656/) (2021)
- [An Experimental Evaluation of Covariates Effects on Unconstrained Face Verification](https://mlsanity.com/paper/1808.05508/) (2018)

## Built on by (most cited first, 25 of 25)

- [Fairness and Bias Mitigation in Computer Vision: A Survey](https://mlsanity.com/paper/2408.02464/) (2024)
- [Synthetic Face Datasets Generation via Latent Space Exploration from Brownian Identity Diffusion](https://mlsanity.com/paper/2405.00228/) (2024)
- [Social Perception of Faces in a Vision-Language Model](https://mlsanity.com/paper/2408.14435/) (2024)
- [Exploring Social Bias in Downstream Applications of Text-to-Image Foundation Models](https://mlsanity.com/paper/2312.10065/) (2023)
- [Review of Demographic Fairness in Face Recognition](https://mlsanity.com/paper/2502.02309/) (2025)
- [FastAvatar: Instant 3D Gaussian Splatting for Faces from Single Unconstrained Poses](https://mlsanity.com/paper/2508.18389/) (2025)
- [Person-Centric Annotations of LAION-400M: Auditing Bias and Its Transfer to Models](https://mlsanity.com/paper/2510.03721/) (2025)
- [Attention IoU: Examining Biases in CelebA using Attention Maps](https://mlsanity.com/paper/2503.19846/) (2025)
- [Unraveling Machine Behavior by Multi-Level Bias Analysis and Detection: Methodology and Application to Computer Vision](https://mlsanity.com/paper/2607.07236/) (2026)
- [CounterFace: A Synthetic Face Dataset for Fine-Grained Counterfactual Evaluation of Face Recognition Systems](https://mlsanity.com/paper/2407.13922/) (2024)
- [Explaining How Quantization Disparately Skews a Model](https://mlsanity.com/paper/2509.07222/) (2025)
- [Discovering Intersectional Bias via Directional Alignment in Face Recognition Embeddings](https://mlsanity.com/paper/2510.15520/) (2025)
- [Automated Detection of Visual Attribute Reliance with a Self-Reflective Agent](https://mlsanity.com/paper/2510.21704/) (2025)
- [What's color got to do with it? Face recognition in grayscale](https://mlsanity.com/paper/2309.05180/) (2023)
- [GELDA: A generative language annotation framework to reveal visual biases in datasets](https://mlsanity.com/paper/2311.18064/) (2023)
- [Are Pose Estimators Ready for the Open World? STAGE: A GenAI Toolkit for Auditing 3D Human Pose Estimators](https://mlsanity.com/paper/2408.16536/) (2024)
- [A Rapid Test for Accuracy and Bias of Face Recognition Technology](https://mlsanity.com/paper/2502.14996/) (2025)
- [Accuracy and Fairness of Facial Recognition Technology in Low-Quality Police Images: An Experiment With Synthetic Faces](https://mlsanity.com/paper/2505.14320/) (2025)
- [Bias at the End of the Score](https://mlsanity.com/paper/2604.13305/) (2026)
- [Identifying Ethical Biases in Action Recognition Models](https://mlsanity.com/paper/2604.17971/) (2026)
- [Do Counterfactually Fair Image Classifiers Satisfy Group Fairness? -- A Theoretical and Empirical Study](https://mlsanity.com/paper/2607.06603/) (2026)
- [Skin Tone in the Wild: A Dataset and Benchmark for 10-tone MST Classification](https://mlsanity.com/paper/2603.02475/) (2026)
- [Splatshot: 3D Face Avatar Generation from a Single Unconstrained Photo](https://mlsanity.com/paper/2606.01493/) (2026)
- [Learning Late, Guiding Early: Timestep-Decoupled Semantic Guidance for Fair Face Generation](https://mlsanity.com/paper/2608.25862/) (2026)
- [DenseFace: Bias Mitigation in Face Recognition via Density-Aware Probabilistic Matching](https://mlsanity.com/paper/2609.16149/) (2026)

## Abstract (arXiv)

We propose an experimental method for measuring bias in face recognition systems. Existing methods to measure bias depend on benchmark datasets that are collected in the wild and annotated for protected (e.g., race, gender) and non-protected (e.g., pose, lighting) attributes. Such observational datasets only permit correlational conclusions, e.g., "Algorithm A's accuracy is different on female and male faces in dataset X.". By contrast, experimental methods manipulate attributes individually and thus permit causal conclusions, e.g., "Algorithm A's accuracy is affected by gender and skin color." Our method is based on generating synthetic faces using a neural face generator, where each attribute of interest is modified independently while leaving all other attributes constant. Human observers crucially provide the ground truth on perceptual identity similarity between synthetic image pairs. We validate our method quantitatively by evaluating race and gender biases of three research-grade face recognition models. Our synthetic pipeline reveals that for these algorithms, accuracy is lower for Black and East Asian population subgroups. Our method can also quantify how perceptual changes in attributes affect face identity distances reported by these models. Our large synthetic dataset, consisting of 48,000 synthetic face image pairs (10,200 unique synthetic faces) and 555,000 human annotations (individual attributes and pairwise identity comparisons) is available to researchers in this important area.