How Do Diffusion Classifiers Decide? A Bias-Centric Evaluation

• URL: http://arxiv.org/abs/2607.03831v1
• Date: Sat, 04 Jul 2026 11:35:04 GMT
• Title: How Do Diffusion Classifiers Decide? A Bias-Centric Evaluation
• Abstract summary: We introduce ASOB-Bench, a bias evaluation for diffusion classifiers along three dimensions: Attribute binding, Size-Order bias, and Background dependency.<n>These dimensions serve not as an exhaustive taxonomy but as targeted probes of how the text-conditioned reconstruction-error score reaches a decision.<n>We find diffusion classifiers are less prone to attribute misbinding than an OpenCLIP baseline; on the established ComCo benchmark they are substantially more susceptible to size-order shortcuts.<n>On ImageNet-B they suffer far larger accuracy drops, revealing heavy reliance on background over foreground cues.
• Score: 2.557366331463333
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Diffusion models have recently been repurposed for zero-shot classification, giving rise to diffusion classifiers that identify the best-matching text prompt by minimizing the noise-prediction error. Despite their growing adoption, how these models make classification decisions remains poorly understood. We introduce ASOB-Bench, a bias evaluation for diffusion classifiers along three dimensions: Attribute binding, Size-Order bias, and Background dependency. These dimensions serve not as an exhaustive taxonomy but as targeted probes of how the text-conditioned reconstruction-error score reaches a decision. Such a perspective is well studied for discriminative vision-language models, yet remains overlooked for diffusion classifiers. Extending an existing framework with five new attribute categories on newly constructed datasets, we find diffusion classifiers are less prone to attribute misbinding than an OpenCLIP baseline; on the established ComCo benchmark they are substantially more susceptible to size-order shortcuts; and on ImageNet-B they suffer far larger accuracy drops, revealing heavy reliance on background over foreground cues. Reconstruction-error heatmaps and U-Net cross-attention visualizations expose the mechanism behind each bias. Because diffusion classifiers share the same denoiser as text-to-image models, these single-pass diagnostics also point toward analogous failure modes in generation. Overall, diffusion classifiers exhibit a distinct bias profile from vision-language models, offering guidance for building more robust diffusion-based models.

Related papers

• Revisiting Autoregressive Models for Generative Image Classification [24.39392782908908]
  We revisit visual AR-based generative classifiers and identify an important limitation of prior approaches.<n>We observe that single-order predictions rely more on partial discriminative cues, while averaging over multiple token orders provides a more comprehensive signal.<n>We leverage recent any-order AR models to estimate order-marginalized predictions, unlocking the high classification potential of AR models.
  arXiv  Detail & Related papers  (2026-03-19T16:44:23Z)
• Generative Classifiers Avoid Shortcut Solutions [84.23247217037134]
  Discriminative approaches to classification often learn shortcuts that hold in-distribution but fail under minor distribution shift.<n>We show that generative classifiers can avoid this issue by modeling all features, both core and spurious, instead of mainly spurious ones.<n>We find that diffusion-based and autorerimigressive generative classifiers achieve state-of-the-art performance on five standard image and text distribution shift benchmarks.
  arXiv  Detail & Related papers  (2025-12-31T18:31:46Z)
• SCALEX: Scalable Concept and Latent Exploration for Diffusion Models [59.86284983662119]
  Image generation models frequently encode social biases, including stereotypes tied to gender, race, and profession.<n>We introduce SCALEX, a framework for scalable and automated exploration of diffusion model latent spaces.<n>It extracts semantically meaningful directions from H-space using only natural language prompts, enabling zero-shot interpretation without retraining or labelling.
  arXiv  Detail & Related papers  (2025-11-13T22:02:44Z)
• Selective Underfitting in Diffusion Models [62.51850508420274]
  We argue that diffusion models underfit the empirical score due to training-time inductive biases.<n>Instead of underfitting the score everywhere, better diffusion models more accurately approximate the score in certain regions of input space, while underfitting it in others.<n>Our results establish that selective underfitting is essential for understanding diffusion models, yielding new, testable insights into their generalization and generative performance.
  arXiv  Detail & Related papers  (2025-10-01T19:06:20Z)
• Studying Classifier(-Free) Guidance From a Classifier-Centric Perspective [100.54185280153753]
  We find that both classifier guidance and classifier-free guidance achieve conditional generation by pushing the denoising diffusion trajectories away from decision boundaries.<n>We propose a generic postprocessing step built upon flow-matching to shrink the gap between the learned distribution for a pretrained denoising diffusion model and the real data distribution.
  arXiv  Detail & Related papers  (2025-03-13T17:59:59Z)
• Generalized Interpolating Discrete Diffusion [65.74168524007484]
  Masked diffusion is a popular choice due to its simplicity and effectiveness.<n>We generalize a new family of general interpolating discrete diffusion (GIDD) which offers greater flexibility in the design of the noising processes.<n>Exploiting GIDD's flexibility, we explore a hybrid approach combining masking and uniform noise, leading to improved sample quality.
  arXiv  Detail & Related papers  (2025-03-06T14:30:55Z)
• Conditional Diffusion Models are Medical Image Classifiers that Provide Explainability and Uncertainty for Free [0.7624308578421438]
  This work presents the first exploration of the potential of class conditional diffusion models for 2D medical image classification.<n>We develop a novel majority voting scheme shown to improve the performance of medical diffusion classifiers.<n>Experiments on the CheXpert and ISIC Melanoma skin cancer datasets demonstrate that foundation and trained-from-scratch diffusion models achieve competitive performance.
  arXiv  Detail & Related papers  (2025-02-06T00:37:21Z)
• Debiasing Classifiers by Amplifying Bias with Latent Diffusion and Large Language Models [9.801159950963306]
  We introduce DiffuBias, a novel pipeline for text-to-image generation that enhances classifier robustness by generating bias-conflict samples. DrouBias is the first approach leveraging a stable diffusion model to generate bias-conflict samples in debiasing tasks. Our comprehensive experimental evaluations demonstrate that DiffuBias achieves state-of-the-art performance on benchmark datasets.
  arXiv  Detail & Related papers  (2024-11-25T04:11:16Z)
• Your Diffusion Model is Secretly a Certifiably Robust Classifier [40.31532959207627]
  We prove that diffusion classifiers possess $O(1)$ Lipschitzness, and establish their certified robustness, demonstrating their inherent resilience.<n>Results show over 80% and 70% certified robustness on CIFAR-10 under adversarial perturbations with (ell) norms less than 0.25 and 0.5, respectively.
  arXiv  Detail & Related papers  (2024-02-04T02:09:18Z)
• Your Diffusion Model is Secretly a Zero-Shot Classifier [90.40799216880342]
  We show that density estimates from large-scale text-to-image diffusion models can be leveraged to perform zero-shot classification. Our generative approach to classification attains strong results on a variety of benchmarks. Our results are a step toward using generative over discriminative models for downstream tasks.
  arXiv  Detail & Related papers  (2023-03-28T17:59:56Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.