Do Hierarchical Networks Actually Learn? Insights from Word-of-Mouth Estimation

We study recursive estimation of a constant state vector by a hierarchical network of asymmetrically informed linear estimators, modeled as a directed tree. Only the leaves observe the regression signal, whereas every other estimator learns by word of mouth from noisy signal estimates relayed by its children. Authority within the hierarchy is expressed through common prior information: after each round, every estimator adopts the estimate and error covariance broadcast by the root. The more accurate this common information, the larger the effective noise variance on local observations, which makes estimation intrinsically self-defeating. The resulting performance is the subject of this paper. We show that recursive least-squares (RLS) estimators do learn the state in mean square, albeit slowly. Under strong persistency of excitation, the broadcast error covariance decays polynomially in time, at a rate that we characterize by upper and lower bounds of matching order. The exponent is determined by the smallest number of relays between a leaf and the root, and it decreases exponentially with this number.

Authors

Publication Details

Journal
arXiv (Cornell University)
Published
2026-11-16
DOI
https://doi.org/10.48550/arxiv.2611.20806
Primary Topic
Control Systems and Identification
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Do Hierarchical Networks Actually Learn? Insights from Word-of-Mouth Estimation

Cristian R. Rojas, Andrea Da Col
arXiv (Cornell University)
Control Systems and Identification
preprint

Do Hierarchical Networks Actually Learn? Insights from Word-of-Mouth Estimation

Cristian R. Rojas, Andrea Da Col
preprint en

Abstract

We study recursive estimation of a constant state vector by a hierarchical network of asymmetrically informed linear estimators, modeled as a directed tree. Only the leaves observe the regression signal, whereas every other estimator learns by word of mouth from noisy signal estimates relayed by its children. Authority within the hierarchy is expressed through common prior information: after each round, every estimator adopts the estimate and error covariance broadcast by the root. The more accurate this common information, the larger the effective noise variance on local observations, which makes estimation intrinsically self-defeating. The resulting performance is the subject of this paper. We show that recursive least-squares (RLS) estimators do learn the state in mean square, albeit slowly. Under strong persistency of excitation, the broadcast error covariance decays polynomially in time, at a rate that we characterize by upper and lower bounds of matching order. The exponent is determined by the smallest number of relays between a leaf and the root, and it decreases exponentially with this number.

arXiv (Cornell University)
Control Systems and Identification
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon