Faithful Embeddings of Irregular and Asynchronous Data for Online Log-NCDEs

2026-05-28 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

A new approach introduces faithful embeddings for Online Log-NCDEs, a universal class of continuous-time models designed for irregular and asynchronous data. This method addresses the sensitivity of traditional interpolation- and imputation-based embeddings by demonstrating that explicit data reconstruction is unnecessary. The research shows that compact-set universality transfers from the model input space to the data space given a continuous and injective embedding. The proposed technique records observations as increments, composing them over query intervals to directly form log-signatures, which offers interval-level summaries without prior variable interpolation and supports online computation. Experiments on synthetic controlled dynamics and real-world time-series datasets confirm the representation's accuracy, efficiency, and robustness to irregular, asynchronous, and sparse observations.

Key takeaway

For Machine Learning Engineers developing continuous-time models with irregular or asynchronous time-series data, you should consider adopting Log-NCDEs with this new faithful embedding. This method eliminates the need for sensitive data interpolation, offering a more robust and efficient way to process sparse observations by directly forming log-signatures from increments. Implementing this approach can lead to more accurate models and streamline online computation for your applications.

Key insights

Explicit data reconstruction is unnecessary for continuous-time models when using a continuous and injective embedding.

Principles

• Compact-set universality transfers from model input to data space with continuous, injective embeddings.

Method

Records observations as increments, composing them over arbitrary query intervals to directly form log-signatures for interval-level summaries, supporting online computation.

In practice

• Provides accurate, efficient, robust representations for irregular, asynchronous, sparse time-series data.
• Applicable to synthetic controlled dynamics and real-world datasets.

Topics

• Neural Controlled Differential Equations
• Log-NCDEs
• Time-series analysis
• Irregular data
• Asynchronous data
• Data embeddings

Best for: Research Scientist, AI Scientist, Machine Learning Engineer

Related on AIssential

• Observable Neural ODEs for Identifiable Causal Forecasting in Continuous Time — Observability of latent states is crucial for identifying dynamic treatment effects in continuous-time causal forecasting with hidden confounders.
• Rescaled Asynchronous SGD: Optimal Distributed Optimization under Data and System Heterogeneity — Rescaled ASGD uses worker-specific stepsize rescaling to correct objective bias in heterogeneous asynchronous distributed learning.
• Statistical Embeddings for Similarity, Retrieval, and Interpretable Alignment of Numeric Tabular Datasets — Statistical embeddings enable interpretable similarity and alignment of heterogeneous numeric tabular datasets for LLMs, preserving statistical context.
• Conformal Prediction with Time-Series Data via Sequential Conformalized Density Regions — SCDR offers doubly robust, asymptotically valid conformal prediction for time-series, generating informative, potentially disconnected prediction sets.
• Multi-Level Causal Embeddings — Causal embeddings generalize abstraction, mapping multiple detailed models into a coarser causal model's sub-systems.
• Dilated CNNs for Periodic Signal Processing: A Low-Complexity Approach — R-DCNN offers efficient periodic signal denoising and waveform estimation using DCNNs and re-sampling for resource-constrained settings.

Open in AIssential →

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