End-to-End Deep Learning for Predicting Metric Space-Valued Outputs

2025-12-31 · Source: JMLR · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Mathematics & Computational Sciences · Depth: Expert, quick

Summary

The E2M (End-to-End Metric regression) framework, introduced by Yidong Zhou, Su I Iao, and Hans-Georg Müller in 2026, provides a deep learning solution for predicting structured, non-Euclidean outputs. These outputs, such as probability distributions, networks, and symmetric positive-definite matrices, are naturally modeled as elements of general metric spaces, where traditional regression fails. E2M operates by performing prediction via weighted Fréchet means over training outputs, with a neural network learning the necessary weights conditioned on the input. This approach ensures geometry-aware prediction, bypassing surrogate embeddings and restrictive parametric assumptions while fully preserving the output space's intrinsic geometry. The framework includes theoretical guarantees, such as a universal approximation theorem and convergence analysis. Simulations demonstrate E2M consistently achieves state-of-the-art performance, particularly with larger sample sizes, and its practical utility is shown in applications like human mortality distributions and New York City taxi networks.

Key takeaway

For research scientists developing models for complex, non-Euclidean data, E2M offers a robust deep learning framework. You should consider E2M to directly predict metric space-valued outputs like probability distributions or networks, avoiding restrictive embeddings. Its theoretical guarantees and demonstrated state-of-the-art performance, especially with larger datasets, suggest it can improve model accuracy and geometric fidelity in your applications. Explore the provided code for implementation.

Key insights

E2M is a deep learning framework for geometry-aware prediction of non-Euclidean, metric space-valued outputs.

Principles

• Prediction via weighted Fréchet means.
• Neural networks learn output weights.
• Preserves intrinsic output geometry.

Method

E2M uses a neural network to learn weights for Fréchet means, applied to training outputs, enabling direct prediction in metric spaces without surrogate embeddings.

In practice

• Predict human mortality distributions.
• Analyze New York City taxi networks.
• Handle symmetric positive-definite matrices.

Topics

• Deep Learning
• Metric Spaces
• Fréchet Mean
• Non-Euclidean Data
• Structured Prediction
• Neural Networks
• Regression

Code references

• SUIIAO/E2M

Best for: AI Scientist, Research Scientist

Related on AIssential

• Neural Networks on Symmetric Spaces of Noncompact Type — A unified point-to-hyperplane distance enables novel neural network architectures on noncompact symmetric spaces.
• Enhancing Safety of Large Language Models via Embedding Space Separation — Explicitly separating harmful and harmless embeddings in LLMs enhances safety without sacrificing general capabilities.
• The Data Manifold under the Microscope — A new framework provides ground-truth geometric data for validating deep learning theory and estimators.
• S2MAM: Semi-supervised Meta Additive Model for Robust Estimation and Variable Selection — S²MAM uses bilevel optimization for robust semi-supervised learning, identifying variables and updating similarity for interpretable predictions.
• Improved Knowledge Distillation for Land-Use Image Classification — Combining hard and soft supervision with KL divergence and Cosine Similarity improves knowledge distillation for land-use classification.
• Structure Retention in Embedding Spaces as a Predictor of Benchmark Performance — Embedding space organization, specifically nearest-neighbor overlap and ICA magnitude, strongly predicts model performance.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by JMLR.