Guided Manifold Alignment with Geometry-Regularized Twin Autoencoders

2025-09-26

AI Summary

Traditional manifold alignment (MA) methods lack out-of-sample generalization capability, limiting their applicability in real-world cross-domain scenarios. To address this, we propose a geometrically regularized dual-branch autoencoder framework that jointly optimizes embedding conformality, cross-modal structural consistency, and reconstruction fidelity via pre-trained alignment guidance and multi-task learning. Crucially, we embed manifold geometric constraints directly into the autoencoder’s latent space, enabling zero-shot out-of-sample extension without retraining. Extensive experiments on multiple benchmark datasets demonstrate significant improvements in cross-domain embedding consistency, information preservation, and transfer performance. In a multimodal Alzheimer’s disease diagnosis task—integrating PET, MRI, and clinical data—our method achieves a 4.2% absolute gain in prediction accuracy, validating its strong generalizability and clinical utility.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Multi-modal Vision
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Manifold alignment (MA) involves a set of techniques for learning shared representations across domains, yet many traditional MA methods are incapable of performing out-of-sample extension, limiting their real-world applicability. We propose a guided representation learning framework leveraging a geometry-regularized twin autoencoder (AE) architecture to enhance MA while enabling generalization to unseen data. Our method enforces structured cross-modal mappings to maintain geometric fidelity in learned embeddings. By incorporating a pre-trained alignment model and a multitask learning formulation, we improve cross-domain generalization and representation robustness while maintaining alignment fidelity. We evaluate our approach using several MA methods, showing improvements in embedding consistency, information preservation, and cross-domain transfer. Additionally, we apply our framework to Alzheimer's disease diagnosis, demonstrating its ability to integrate multi-modal patient data and enhance predictive accuracy in cases limited to a single domain by leveraging insights from the multi-modal problem.

Problem

• Enabling out-of-sample generalization for manifold alignment methods
• Maintaining geometric fidelity in cross-domain embedding mappings
• Improving multi-modal data integration for clinical diagnosis applications

Innovation

• Geometry-regularized twin autoencoders enable manifold alignment
• Pre-trained alignment model improves cross-domain generalization
• Multitask learning maintains geometric fidelity in embeddings