LIVS: A Pluralistic Alignment Dataset for Inclusive Public Spaces

2025-02-27

AI Summary

Existing multi-criteria-aligned text-to-image (T2I) models struggle to accommodate context-sensitive community needs in inclusive urban planning. Method: We propose an intersectionality-informed, community-preference-driven alignment paradigm and introduce LIVS—the first domain-specific multi-criteria T2I benchmark—comprising 37,000 preference pairs across six spatial value dimensions, co-developed over two years with 30 community organizations. Participatory design is systematically integrated into both data curation and evaluation, challenging the assumption of technical neutrality. We fine-tune Stable Diffusion XL using Direct Preference Optimization (DPO) and incorporate a novel multidimensional value encoding framework. Contribution/Results: Experiments demonstrate significantly improved alignment with community preferences; performance scales positively with preference dataset size; and the high prevalence of neutral preference judgments underscores the intrinsic complexity of intersectional modeling.

Technology Category

• Humans and AI: Learning Human Values and Preferences
• Knowledge Representation and Reasoning: Preferences
• Machine Learning: Learning Preferences or Rankings

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

We introduce the Local Intersectional Visual Spaces (LIVS) dataset, a benchmark for multi-criteria alignment of text-to-image (T2I) models in inclusive urban planning. Developed through a two-year participatory process with 30 community organizations, LIVS encodes diverse spatial preferences across 634 initial concepts, consolidated into six core criteria: Accessibility, Safety, Comfort, Invitingness, Inclusivity, and Diversity, through 37,710 pairwise comparisons. Using Direct Preference Optimization (DPO) to fine-tune Stable Diffusion XL, we observed a measurable increase in alignment with community preferences, though a significant proportion of neutral ratings highlights the complexity of modeling intersectional needs. Additionally, as annotation volume increases, accuracy shifts further toward the DPO-tuned model, suggesting that larger-scale preference data enhances fine-tuning effectiveness. LIVS underscores the necessity of integrating context-specific, stakeholder-driven criteria into generative modeling and provides a resource for evaluating AI alignment methodologies across diverse socio-spatial contexts.

Problem

• Develops LIVS dataset for inclusive urban planning alignment.
• Fine-tunes T2I models using community-driven criteria.
• Evaluates AI alignment in diverse socio-spatial contexts.

Innovation

• Developed LIVS dataset for inclusive urban planning.
• Used Direct Preference Optimization for model fine-tuning.
• Enhanced alignment with community-driven spatial preferences.