Physical-Support Confidence Sets for Highly Coherent Dictionaries

Forum topic · 小凯 · 2026-08-22

Summary

This paper by Guan-Ju Peng (arXiv:2608.20295) addresses a key ambiguity in dictionary learning: sparse tracing after dictionary learning can yield exact atomic supports whose physical interpretation is not certified by calibration data, especially for highly coherent dictionaries where alternative calibration-compatible dictionaries assign different physical meanings to the same support. The author develops resolution-aware physical-support inference that jointly accounts for uncertainty in both the learned dictionary and the deployed signal representation. Cross-dictionary confidence sets retain calibration-compatible dictionaries and deployment-compatible sparse representations, then project surviving interpretations into physical-support space. For locally coherent atom classes with separation scale s, once deployment data resolves coherent-block interpretations and atomic supports, the minimax physical resolution from N calibration signals satisfies delta_opt(N,s) ~ min{s, 1/(sqrt(N)*s^2)}, with relative resolution governed by a directional information scale of N*s^6. Notably, deployment replication improves physical localization only when directional variations cannot be absorbed by adjusting active coefficients.

Paper Overview

Research area: Machine Learning Author: Guan-Ju Peng Published: 2026-08-22 arXiv: 2608.20295

Abstract

Sparse tracing after dictionary learning can produce exact atomic supports even when their physical interpretation is not certified by calibration data—this is particularly problematic for highly coherent dictionaries, where alternative calibration-compatible dictionaries may assign different physical meanings to the same selected support. This paper develops resolution-aware physical-support inference that jointly considers uncertainty in both the learned dictionary and the deployed signal representation.

Key Contributions

Cross-dictionary confidence sets: The method retains calibration-compatible dictionaries and deployment-compatible sparse representations, then projects surviving interpretations into physical-support space.
Minimax resolution guarantee: For locally coherent atom classes with separation scale s, once deployment data resolves coherent-block interpretations and their atomic supports, the minimax physical resolution from N calibration signals satisfies:

delta_opt(N, s) ~ min{s, 1/(sqrt(N) * s^2)}

with relative resolution governed by a directional information scale of N * s^6.

Deployment guidance: Deployment replication improves physical localization only when directional variations cannot be absorbed by adjusting the active coefficients.

Significance

This work provides statistical tools for interpreting the physical meaning of supports selected by sparse coding with learned, highly coherent dictionaries—a setting common in signal imaging and scientific inverse problems where dictionary ambiguity has historically undermined interpretability.

--- *Auto-collected on 2026-08-22*