Research map / Probabilistic and scientific modeling / Molecules and proteins

Harnessing Data Asymmetry in Manifold Learning

NeurIPS 2026 · Probabilistic and scientific modeling › Molecules and proteins · approach: Equivariant and disentangled latent representations

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes exploiting asymmetric dissimilarities in manifold learning, moving beyond symmetric Riemannian geometry to preserve directed pairwise relations in low-dimensional embeddings.

Problem

Standard manifold learning relies on symmetric Riemannian geometry, forcing symmetric dissimilarities and embedding spaces (e.g. Euclidean), which discards valuable asymmetric information present in real high-dimensional data.

Approach

Treats asymmetric pairwise dissimilarities as first-class in manifold learning, so embeddings preserve directional data relations rather than symmetrizing them; the abstract indicates this also relaxes the symmetric embedding space assumption.

Closest papers by meaning

• Finsler Multi-Dimensional Scaling: Manifold Learning for Asymmetric Dimensionality Reduction and Embedding (CVPR 2025 · similarity 0.95)
• Learning Mixed-Curvature Representations in Product Spaces (ICLR 2019 · similarity 0.88)
• Score-based Pullback Riemannian Geometry: Extracting the Data Manifold Geometry using Anisotropic Flows (ICML 2025 · similarity 0.88)
• Euclidean Embedding of Data Using Local Distances (NeurIPS 2026 · similarity 0.86)
• Graph Geometry-Preserving Autoencoders (ICML 2024 · similarity 0.86)
• Partition First, Embed Later: Laplacian-Based Feature Partitioning for Refined Embedding and Visualization of High-Dimensional Data (ICML 2025 · similarity 0.86)