Reliable Semantic Transport with Function Preservation for Label-Efficient Federated Load Event Recognition

Abstract Federated non-intrusive load monitoring (NILM) can exploit appliance-related knowledge across households without centralizing raw measurements. Its prac tical use is limited by sparse local labels and household-specific operating patterns, which make cross-client semantic evidence unevenly reliable. We propose FedReST, a reliable semantic transport framework for label-efficient federated load event recognition. FedReST organizes collaboration as a two admission process: the server assesses whether semantic evidence is sufficiently supported for sharing, and each receiving client assesses whether the resulting direction is compatible with its local function. Masked reconstruction and state aware contrastive learning construct temporal representations that are mapped to simplex-valued semantic embeddings. Clients transmit evidence-weighted task state prototype measures, which are aligned through reliability-gated partial optimal transport so that unsupported mass can remain unmatched. A function preserving gradient projection suppresses shared-update components associated with locally anchor-sensitive parameter directions. On frozen REFIT 2014 rolling folds with matched two-epoch local budgets and 10% visible labels, corrected-core FedReST obtains higher mean Event F1 and Window F1 than an architecture matched Head+BN FedAvg reference, with paired differences of +0.00524 and +0.00104, respectively. It also improves mean Event F1 by +0.01590 over equal client FedAvg. The results support reliability-aware semantic sharing as a useful design principle for heterogeneous, label-efficient federated load recognition.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-16
DOI
https://doi.org/10.21203/rs.3.rs-11548524/v1
Primary Topic
Smart Grid Energy Management
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Reliable Semantic Transport with Function Preservation for Label-Efficient Federated Load Event Recognition

Jiao Tian, Yixin Zhao
3 links
Research Square
Smart Grid Energy Management
preprint

Reliable Semantic Transport with Function Preservation for Label-Efficient Federated Load Event Recognition

Jiao Tian, Yixin Zhao
preprint en

Abstract

Abstract Federated non-intrusive load monitoring (NILM) can exploit appliance-related knowledge across households without centralizing raw measurements. Its prac tical use is limited by sparse local labels and household-specific operating patterns, which make cross-client semantic evidence unevenly reliable. We propose FedReST, a reliable semantic transport framework for label-efficient federated load event recognition. FedReST organizes collaboration as a two admission process: the server assesses whether semantic evidence is sufficiently supported for sharing, and each receiving client assesses whether the resulting direction is compatible with its local function. Masked reconstruction and state aware contrastive learning construct temporal representations that are mapped to simplex-valued semantic embeddings. Clients transmit evidence-weighted task state prototype measures, which are aligned through reliability-gated partial optimal transport so that unsupported mass can remain unmatched. A function preserving gradient projection suppresses shared-update components associated with locally anchor-sensitive parameter directions. On frozen REFIT 2014 rolling folds with matched two-epoch local budgets and 10% visible labels, corrected-core FedReST obtains higher mean Event F1 and Window F1 than an architecture matched Head+BN FedAvg reference, with paired differences of +0.00524 and +0.00104, respectively. It also improves mean Event F1 by +0.01590 over equal client FedAvg. The results support reliability-aware semantic sharing as a useful design principle for heterogeneous, label-efficient federated load recognition.

Research Square
Xinjiang University (CN)
Smart Grid Energy Management
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