# Unraveling Machine Behavior by Multi-Level Bias Analysis and Detection: Methodology and Application to Computer Vision

arXiv: https://arxiv.org/abs/2607.07236 · HTML: https://mlsanity.com/paper/2607.07236/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Builds on 29 of its 45 references that are in the map.
- First posted on arXiv 8 Jul 2026.

## Builds on (most cited first, 29 of 29)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [Intriguing properties of neural networks](https://mlsanity.com/paper/1312.6199/) (2013)
- [Visualizing and Understanding Convolutional Networks](https://mlsanity.com/paper/1311.2901/) (2013)
- [Shortcut Learning in Deep Neural Networks](https://mlsanity.com/paper/2004.07780/) (2020)
- [A Survey on Bias and Fairness in Machine Learning](https://mlsanity.com/paper/1908.09635/) (2019)
- [VGGFace2: A dataset for recognising faces across pose and age](https://mlsanity.com/paper/1710.08092/) (2017)
- [Last Layer Re-Training is Sufficient for Robustness to Spurious Correlations](https://mlsanity.com/paper/2204.02937/) (2022)
- [Fairness in Machine Learning: A Survey](https://mlsanity.com/paper/2010.04053/) (2020)
- [An Investigation of Why Overparameterization Exacerbates Spurious Correlations](https://mlsanity.com/paper/2005.04345/) (2020)
- [Understanding the Role of Individual Units in a Deep Neural Network](https://mlsanity.com/paper/2009.05041/) (2020)
- [Flexibly Fair Representation Learning by Disentanglement](https://mlsanity.com/paper/1906.02589/) (2019)
- [REVISE: A Tool for Measuring and Mitigating Bias in Visual Datasets](https://mlsanity.com/paper/2004.07999/) (2020)
- [Bias Mitigation for Machine Learning Classifiers: A Comprehensive Survey](https://mlsanity.com/paper/2207.07068/) (2022)
- [Turning a Blind Eye: Explicit Removal of Biases and Variation from Deep Neural Network Embeddings](https://mlsanity.com/paper/1809.02169/) (2018)
- [A Comprehensive Study on Face Recognition Biases Beyond Demographics](https://mlsanity.com/paper/2103.01592/) (2021)
- [Spurious Features Everywhere -- Large-Scale Detection of Harmful Spurious Features in ImageNet](https://mlsanity.com/paper/2212.04871/) (2022)
- [InsideBias: Measuring Bias in Deep Networks and Application to Face Gender Biometrics](https://mlsanity.com/paper/2004.06592/) (2020)
- [Examining CNN Representations with respect to Dataset Bias](https://mlsanity.com/paper/1710.10577/) (2017)
- [Benchmarking Algorithmic Bias in Face Recognition: An Experimental Approach Using Synthetic Faces and Human Evaluation](https://mlsanity.com/paper/2308.05441/) (2023)
- [Algorithmic Discrimination: Formulation and Exploration in Deep Learning-based Face Biometrics](https://mlsanity.com/paper/1912.01842/) (2019)
- [SensitiveNets: Learning Agnostic Representations with Application to Face Images](https://mlsanity.com/paper/1902.00334/) (2019)
- [Second FRCSyn-onGoing: Winning Solutions and Post-Challenge Analysis to Improve Face Recognition with Synthetic Data](https://mlsanity.com/paper/2412.01383/) (2024)
- [Addressing Bias in LLMs: Strategies and Application to Fair AI-based Recruitment](https://mlsanity.com/paper/2506.11880/) (2025)
- [IFBiD: Inference-Free Bias Detection](https://mlsanity.com/paper/2109.04374/) (2021)
- [Measuring Bias in AI Models: An Statistical Approach Introducing N-Sigma](https://mlsanity.com/paper/2304.13680/) (2023)
- [FairCVtest Demo: Understanding Bias in Multimodal Learning with a Testbed in Fair Automatic Recruitment](https://mlsanity.com/paper/2009.07025/) (2020)
- [Towards Measuring Bias in Image Classification](https://mlsanity.com/paper/2107.00360/) (2021)
- [Balancing Tails when Comparing Distributions: Comprehensive Equity Index (CEI) with Application to Bias Evaluation in Operational Face Biometrics](https://mlsanity.com/paper/2506.10564/) (2025)
- [Discovering Intersectional Bias via Directional Alignment in Face Recognition Embeddings](https://mlsanity.com/paper/2510.15520/) (2025)

## Built on by (most cited first, 3 of 3)

- [AIriskEval-edu Demo: Auditing of Pedagogical Risks in Educational Explanations](https://mlsanity.com/paper/2607.25634/) (2026)
- [VideoRun2D Demo: Markerless Body Tracking for Biomechanical Analysis of Running](https://mlsanity.com/paper/2608.19480/) (2026)
- [inLoRA-MINT: Internal Representations for Auditing Training Data in Domain-adapted LLMs](https://mlsanity.com/paper/2610.18051/) (2026)

## Abstract (arXiv)

This study investigates the presence and propagation of bias within Neural Networks through a comprehensive multi-level analysis spanning the learned latent space, layer activations, and the network's parameters. Based on this taxonomy, we propose three bias detection approaches: 1) SpaceBias (new method), which characterizes the latent space prior to the final classification layer using neighbor-probability distributions and quantifies bias with the two-sample Kolmogorov-Smirnov test on the per-group distributions. 2) ActivationBias (extension of the existing method InsideBias), which analyzes the activations of neural network filters and quantifies bias via a Mann-Whitney U test, based on the observed fact that underrepresented groups exhibit lower activation levels in the final convolutional layers. 3) WeightBias (extension of the existing method IFBiD), which uses a secondary neural network trained to identify biased patterns directly in the parameters of task-specific models. Unlike conventional methods, which assess neural network outcomes and treat the model as a black box, our proposed techniques provide insight into how biases manifest within the network architecture itself at different levels, offering a more nuanced and detailed understanding. Experiments are conducted on two complementary applications: gender classification in the DiveFace dataset (72,000 face images) and digit classification on a colored-MNIST benchmark with controlled bias severity. In total, more than 127,000 models with varying degrees and types of bias were trained and evaluated. The severity sweep shows that the internal disparity, and with it the detection performance, decreases smoothly as the training distribution approaches balance. The results highlight the importance of methods that provide deeper insight into the behavior of AI models.