An analysis of 26 text-to-image models across 12 aspects using 62 scenarios and 25 metrics.

Discover HEIM’s enhanced metrics, blending human and automated evaluations to assess AI models on fairness, robustness, multilinguality, and more.

xplore 62 diverse scenarios designed to assess text-to-image AI models across 12 key aspects like alignment, originality, and bias.

Discover how HEIM evaluates text-to-image models across 12 crucial aspects, from aesthetics and originality to bias and multilinguality.

HEIM evaluates AI image models using a 4-part framework: aspects, scenarios, adaptations, and metrics.

The HEIM benchmark evaluates 26 text-to-image models across 12 critical aspects, revealing diverse strengths and weaknesses.

This article outlines the human evaluation process for AI-generated images using Amazon Mechanical Turk and compliance with Human Subjects.

Explore the latest advancements in text-to-image models like Stable Diffusion, DALL-E, and Dreamlike Diffusion.

LLMs are as smart—and as biased—as the humans who trained them. Although AI can’t think for itself, we're just beginning to explore the depths of LLM psychology

Dive deep into the challenges of measuring and mitigating bias in machine learning models, the trade-offs between fairness metrics.

Learn techniques for mitigating bias and promoting diversity in AI prompts.

Feasibility of genAI face images to train Face Recognition (FR) models. Article compares performance of FR models trained on real vs synthetic datasets.

Explore different fairness definitions in medical AI, the challenges of bias mitigation, and how these impact diagnosis accuracy and healthcare outcomes.

Discover how positive-sum fairness in medical AI allows for better performance without harming subgroups, challenging traditional fairness models.

Explore experiments comparing four AI models on chest radiographs, using sensitive attributes like race and evaluating fairness and performance using AUROC.

Compare AI models with sensitive attributes in medical imaging, highlighting fairness, performance, and the impact of demographic information on predictions.

Learn about positive-sum fairness, a new framework that balances AI performance improvements without disadvantaging any subgroup in medical applications.

A new study introduces the concept of positive-sum fairness in AI models for medical imaging, suggesting that larger performance gaps aren't always harmful.

Study shows how positive-sum fairness improves lung lesion detection without harming subgroup performance.

New findings show how positive-sum fairness impacts AI model performance, improving fairness without harming accuracy, with mixed results for different models.