# How are attributes expressed in face DCNNs?

arXiv: https://arxiv.org/abs/1910.05657 · HTML: https://mlsanity.com/paper/1910.05657/

## Key facts

- Cited by 13 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 12 Oct 2019.

## Built on by (most cited first, 13 of 13)

- [Biometrics: Trust, but Verify](https://mlsanity.com/paper/2105.06625/) (2021)
- [PASS: Protected Attribute Suppression System for Mitigating Bias in Face Recognition](https://mlsanity.com/paper/2108.03764/) (2021)
- [Gradient Attention Balance Network: Mitigating Face Recognition Racial Bias via Gradient Attention](https://mlsanity.com/paper/2304.02284/) (2023)
- [3D Face Reconstruction: the Road to Forensics](https://mlsanity.com/paper/2309.11357/) (2023)
- [Distill and De-bias: Mitigating Bias in Face Verification using Knowledge Distillation](https://mlsanity.com/paper/2112.09786/) (2021)
- [Gaussian Harmony: Attaining Fairness in Diffusion-based Face Generation Models](https://mlsanity.com/paper/2312.14976/) (2023)
- [An Explainable Model-Agnostic Algorithm for CNN-based Biometrics Verification](https://mlsanity.com/paper/2307.13428/) (2023)
- [Dissecting Human Body Representations in Deep Networks Trained for Person Identification](https://mlsanity.com/paper/2502.15934/) (2025)
- [A Quantitative Evaluation of the Expressivity of BMI, Pose and Gender in Body Embeddings for Recognition and Identification](https://mlsanity.com/paper/2503.06451/) (2025)
- [Encoding of Demographic and Anatomical Information in Chest X-Ray-based Severe Left Ventricular Hypertrophy Classifiers](https://mlsanity.com/paper/2506.03192/) (2025)
- [PrivacyProber: Assessment and Detection of Soft-Biometric Privacy-Enhancing Techniques](https://mlsanity.com/paper/2211.08864/) (2022)
- [Revisiting Marr in Face: The Building of 2D--2.5D--3D Representations in Deep Neural Networks](https://mlsanity.com/paper/2411.16148/) (2024)
- [AttriBE: Quantifying Attribute Expressivity in Body Embeddings for Recognition and Identification](https://mlsanity.com/paper/2604.27218/) (2026)

## Abstract (arXiv)

As deep networks become increasingly accurate at recognizing faces, it is vital to understand how these networks process faces. While these networks are solely trained to recognize identities, they also contain face related information such as sex, age, and pose of the face. The networks are not trained to learn these attributes. We introduce expressivity as a measure of how much a feature vector informs us about an attribute, where a feature vector can be from internal or final layers of a network. Expressivity is computed by a second neural network whose inputs are features and attributes. The output of the second neural network approximates the mutual information between feature vectors and an attribute. We investigate the expressivity for two different deep convolutional neural network (DCNN) architectures: a Resnet-101 and an Inception Resnet v2. In the final fully connected layer of the networks, we found the order of expressivity for facial attributes to be Age > Sex > Yaw. Additionally, we studied the changes in the encoding of facial attributes over training iterations. We found that as training progresses, expressivities of yaw, sex, and age decrease. Our technique can be a tool for investigating the sources of bias in a network and a step towards explaining the network's identity decisions.