Distributed fire-monitoring systems may involve many sensing devices reporting over a shared uplink, even though the operational objective is often to identify regions exhibiting evidence of fire rather than to recover device-specific measurements. We introduce TERRA (Task-Oriented Edge Representation and Regional Aggregation), an edge-AI framework that combines task-oriented local inference, identity-free region-tagged reporting, and permutation-invariant regional evidence aggregation. TERRA supports two reporting modes: TERRA-Score, in which each device communicates a quantized local fire probability, and TERRA-Feature, in which it communicates a compact learned representation for receiver-side aggregation using a permutation-invariant DeepSets model. Using all 15 numerical sensing modalities of the distributed indoor EN54 fire-alarm dataset as a proof-of-concept testbed, we evaluate regional inference for an episode-level fire-scenario target under repetition-disjoint three-fold cross-validation. We study representation size, task-payload quantization, report availability, and robustness to missed and false recovered messages. With nine reports and perfect message recovery, TERRA-Feature achieves mean AUPRC 0.990 using a 24-bit task payload, while TERRA-Score achieves 0.987 using only 6 bits; both attain an MDR of 1.81% at FAR ≤1%. TERRA-Feature achieves slightly higher AUPRC under reliable recovery, whereas the more compact TERRA-Score is less sensitive to increasing false-message contamination.
Figures & tables
Figure 1. TERRA architecture for region-wise inference over an identity-free random access channel. Devices transmit either a quantized local score (TERRA-Score) or a learned representation (TERRA-Feature), together with a region tag and a nonce; recovered reports are grouped by region and aggregated using score averaging or DeepSets, respectively. TERRA architecture with distributed multimodal sensing nodes performing local score or feature extraction, transmitting region-tagged reports over an identity-free URA uplink, and a receiver grouping recovered reports by region for score averaging or permutation-invariant DeepSets aggregation.
Figure 2. Communication–performance tradeoffs under perfect message recovery. (a) Mean regional AUPRC versus task-payload length with all nine reports available; open markers indicate the selected operating points. (b) Mean regional AUPRC versus the number of available reports using 6-bit TERRA-Score, 24-bit TERRA-Feature, and 30-bit Handcrafted-Stats reports. Shaded regions indicate ±1 standard deviation across the three outer folds after averaging report-subset realizations within each fold. Two plots showing regional AUPRC under perfect message recovery. TERRA-Score and TERRA-Feature approach an AUPRC of 0.99 with compact selected payloads, and both improve as the number of available reports increases.
Method
Report dim.
Task bits
AUPRC
MDR@FAR ≤1%
TERRA-Score
1
6
0.9868
1.81%
TERRA-Feature
4
24
0.9896
1.81%
Handcrafted-Stats
30
30
0.9713
11.42%
Table 1. Selected reporting operating points under perfect message recovery with K=9 . Report dimension refers to the transmitted task representation; task bits exclude the region tag and nonce.
Figure 3. Regional fire-detection robustness to message-recovery errors for the selected reporting configurations with nine transmitted reports per region. Panels show mean AUPRC for (a) TERRA-Score, (b) TERRA-Feature, and (c) Handcrafted-Stats as a function of the global missed-detection probability PMD and false-alarm probability PFA . All panels use the same color scale, and contours, where present, indicate AUPRC levels of 0.98, 0.95, 0.90, and 0.80. Metrics are averaged across 10 Monte Carlo recovery realizations within each outer fold, then averaged equally across the three folds. Three heat maps showing regional AUPRC versus missed-detection and false-alarm probabilities for TERRA-Score, TERRA-Feature, and Handcrafted-Stats. Performance decreases with increasing recovery errors, while TERRA-Score is less sensitive to high false-message contamination over much of the evaluated region.
Edge inference runs trained AI models on nearby devices such as phones, cameras, robots, and embedded computers under tight resource and latency limits.
We present a deployed system for on-orbit wildfire detection aboard a nine-satellite commercial thermal infrared constellation, operating under demanding joint constraints: sub-megabyte model footprint, sub-150 ms per-batch TensorRT FP16 inference on an NVIDIA Jetson Xavier NX, and an end-to-end alert pipeline targeting under 10 minutes from satellite overpass to fire event communication. The system operates on uncalibrated mid-wave infrared (MWIR) single-band imagery at 200 m ground sampling distance, where fires frequently appear as sub-pixel or single-pixel thermal anomalies under extreme class imbalance -- challenges not addressed by the contextual thermal-thresholding pipelines (MODIS, VIIRS) that currently dominate operational fire monitoring. We present an empirical study of lightweight dense representation learning for this regime using a proprietary nine-satellite MWIR dataset. We compare dense masked autoencoding (DenseMAE) and a hybrid DenseMAE+EMA (exponential moving average) distillation variant, and evaluate representations via linear probing and full-distribution pixel-level average precision (AP) under extreme class imbalance. DenseMAE pretraining enables compact downstream models on the latency-accuracy Pareto frontier: our fastest SSL-pretrained model achieves 0.640 test AP and 0.69 event-level Fire-F1 with 65.34 ms latency per batch and a 0.52 MB engine, without pruning or compression. The best configuration reaches 0.699 AP and 0.744 Fire-F1 below 1 MB, outperforming a supervised baseline (0.650 AP) under comparable constraints.
Matthias Rötzer, Veronika Pörtge, Martin Ickerott +6
We study the application of federated learning (FL) to indoor fire detection. Such fire-detection systems use edge cameras that record sensitive footage which cannot easily be collected at a central server. Existing federated solutions leave three practical obstacles unaddressed: limited uplink bandwidth, Byzantine (malicious or faulty) clients, and unconditional trust in a single, permanently fixed aggregation server. Our main contributions address all three. In particular, we provide (i) a curated indoor fire-detection dataset assembled from eight public sources; (ii) an edge-deployable detector whose model updates are compressed up to 10 times with only a small loss in balanced accuracy; and (iii) a semi-decentralized Byzantine-robust FL method that combines history-aware aggregation with a rotating coordinator, evicting stealthy attacks that per-round filters miss while removing the fixed-server single point of failure. On the held-out test set the rotating-coordinator method matches its fixed-server counterpart in accuracy and detection speed, and a physically distributed six-node cloud deployment confirms feasibility.
Georgia Argyrou, Aymen Bahroun, Hedi Fendri +1
Department of Computer Science, Aalto University, Espoo, Finland · Kudelski Labs
We propose a momentum-guided federated split distillation framework for personalized, efficient, and autonomous temporal edge intelligence. We introduce TeRR-SAtt, our novel temporal reservoir student attention design that combines fixed reservoir representations, a lightweight temporal student, and personalized output modules. We also present AMGF, our anticipatory momentum-guided fusion mechanism that clusters clients through learning momentum and derives specialized teacher updates. On real-world smart-building data, TeRR-SAtt reduces edge training latency by 65.50%, inference latency by 44.70%, training memory usage by 18.40%, and inference CPU usage by 33.10% over the considered baselines. At the same time, AMGF improves local learning by up to 35.31% in RMSE compared to global updates.