Pith. sign in

REVIEW 2 major objections 4 minor 35 references

High-resolution satellite imagery plus transfer learning can detect schools and cell towers, then use terrain-aware line-of-sight to flag which schools already sit within practical radio reach.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review

2026-07-12 07:29 UTC pith:RVAVEQ5M

load-bearing objection Clean transfer-learning demo that maps schools and towers then feeds them into ITU CPP LOS; detection numbers are solid, but the 9/11 triage is an unvalidated proxy on a 20-school patch. the 2 major comments →

arxiv 2607.02724 v1 pith:RVAVEQ5M submitted 2026-07-02 cs.CV cs.AIcs.CY

Signal from Space: Detecting Schools and Towers to Bridge the Digital Divide

Zakarya Elmimouni , Sandor Farkas , Fares Fourati , Vladimir Daigele , Walid Mathlouthi , Mohamed-Slim Alouini This is my paper
classification cs.CV cs.AIcs.CY
keywords school detectioncell tower detectionsatellite imagerytransfer learningconnectivity mappingobject detectiondigital divideline-of-sight analysis
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Millions of children still attend schools with no internet because planners lack reliable maps of both the schools and the nearby towers that could serve them. This paper shows that ordinary high-resolution satellite photos are enough: a detector pre-trained elsewhere is fine-tuned on a few dozen local examples, an oriented-box detector finds the towers, and a terrain-aware radio-visibility model checks whether a clear path exists between each school and its nearest tower. Schools with a confirmed line-of-sight link inside one kilometre are labelled high-connectivity-potential; the rest are labelled hard-to-connect. The result is an automated priority map that does not depend on sparse or noisy third-party datasets. A sympathetic reader cares because the same recipe can be repeated in any underserved region once a small golden set is annotated, giving connectivity planners a fast first-pass triage from space alone.

Core claim

A purely imagery-driven two-stage pipeline that fine-tunes object detectors for schools and for oriented cell-tower boxes, then feeds the detections into a point-to-point radio-visibility model, can triage schools into high-connectivity-potential versus hard-to-connect categories. On a real rural test region the fine-tuned school detector reaches 0.913 mAP@50 and the tower detector reaches 0.971; of twenty detected schools, nine show confirmed line-of-sight to a tower inside one kilometre while eleven do not.

What carries the argument

Transfer-learned object detection (standard boxes for schools, oriented boxes for slender towers and their shadows) combined with a terrain-aware point-to-point line-of-sight filter inside a fixed 1 km radius. The detectors supply locations; the visibility step turns raw proximity into an engineering-grade connectivity proxy.

Load-bearing premise

That a confirmed line-of-sight path to any detected cell tower inside a fixed one-kilometre circle is a faithful enough stand-in for real-world school connectivity potential.

What would settle it

Independent ground audits of the twenty detected schools showing that most of the nine labelled high-potential actually lack usable service, or that several of the eleven hard-to-connect schools already have service via non-line-of-sight radio or fibre, would overturn the proxy.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • National-scale school and tower inventories can be generated from imagery alone without waiting for dense third-party tabular features.
  • Limited connectivity investment can be directed first toward schools already inside radio reach of existing towers.
  • The same small-golden-set fine-tuning recipe can be repeated for new countries once a few dozen local labels are collected.
  • High-recall detector settings reduce the chance that existing infrastructure is missed and therefore wrongly marks schools as hard-to-connect.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same shadow-aware oriented-box detector could be reused for other slender vertical assets such as water towers or power poles that also cast distinctive shadows.
  • If higher-resolution elevation models become available, the visibility step could rank multiple candidate towers by estimated path loss rather than binary line-of-sight.
  • Periodic re-inference on new satellite revisits would let planners watch connectivity gaps shrink after investments are made.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. The paper proposes a vision-only transfer-learning pipeline that fine-tunes YOLO-based object detectors (USA-pretrained schools; Mozambique-pretrained towers) on small manually curated Lesotho golden sets to detect schools and oriented cell towers from high-resolution satellite imagery. Detected schools are then assessed for connectivity potential by feeding school and tower centroids into the ITU CPP P2P radio-visibility model (geodesic distance + DEM line-of-sight) inside a fixed 1 km radius, producing a binary triage of high-connectivity-potential versus hard-to-connect schools. On a Mafeteng sub-region the pipeline reports school mAP@50 = 0.913 (Table I, Model 3), tower mAP@50 = 0.971, and a 9/11 split among 20 detections, presented as a scalable, auxiliary-data-free prioritization tool for the Giga Initiative.

Significance. If the LOS proxy is accepted as a useful first-pass triage signal, the work is practically valuable for connectivity mapping in data-scarce regions: it cleanly demonstrates efficient domain adaptation with only ~200 school labels and a 179-tile tower set, supplies a transparent three-way ablation (zero-shot / from-scratch / fine-tune), prioritizes high recall for operational safety, and replaces pure planar distance with an engineering-grade CPP P2P model. The reproducible transfer recipe and explicit integration with ITU tooling are concrete strengths that lower the barrier for Giga-style infrastructure screening even when full ground-truth connectivity labels remain unavailable.

major comments (2)
  1. [§IV.C / §V / Abstract] §IV.C and §V define “high connectivity potential” solely as confirmed LOS to any detected tower inside a fixed 1 km radius, then report the 9/11 split as the key operational outcome. No operator records, speed tests, or even a small hand-labeled connectivity sample are used to measure precision/recall of this binary label. The abstract and conclusion still frame the pipeline as enabling “data-driven prioritization,” while §VI correctly calls the output only a triage signal. Either supply an external validation subset or reframe the central claim throughout to match the weaker, proxy-only language of the discussion.
  2. [§V / §VI / Abstract] The entire connectivity demonstration rests on 20 school detections inside a single Mafeteng sub-region and a tower training set of only 179 tiles (n=188 Type-1). While detection metrics on the golden test sets are strong, the paper offers no multi-district results, no uncertainty quantification on the 9/11 counts, and no sensitivity analysis of the free parameters (confidence threshold 0.6, 1 km radius, stride 256). Claims of “large-scale infrastructure mapping” and “strong performance across this region” therefore exceed the evidence presented; either expand the geographic scope or bound the claims more tightly to the demonstrated feasibility study.
minor comments (4)
  1. [Table I] Table I caption and surrounding text refer to “YOLO26n”; clarify the exact Ultralytics version (YOLOv8/v10/v11/v12 lineage) so readers can reproduce the architecture.
  2. [Fig. 1 / §IV.C] Figure 1 caption and §IV.C both mention a “1 km connectivity zone,” yet the text never states whether the radius is geodesic or planar; a single clarifying sentence would remove ambiguity.
  3. Minor typographic inconsistencies appear throughout (e.g., “School-Connectivity,” “cell ular,” mixed capitalization of “Type-1”). A light copy-edit pass would improve readability.
  4. [§III.B / §V] The tower dataset description (§III.B) notes shadows as a second class but does not report per-class AP; adding a short per-class breakdown would help readers judge whether shadow cues dominate performance.

Circularity Check

1 steps flagged

Mild self-citation of authors' prior school detector supplies base weights; Lesotho mAPs and triage counts remain independent empirical outputs of fine-tuning plus a defined proxy.

specific steps
  1. self citation load bearing [§I (Introduction) and §IV.A (School Detection Model Selection)]
    "Our approach builds on a scalable school detection pipeline introduced in [7] ... Finetune The best YOLO26n model pre-trained on the USA golden dataset [30]."

    The load-bearing school detector is initialized from weights produced by the same authors' concurrent paper [7]/30]. While the subsequent fine-tuning and mAP evaluation on Lesotho data are independent, the methodological premise that a high-quality detector already exists for transfer is justified solely by that self-citation rather than by an external, independently verified result.

full rationale

The paper's derivation chain is an engineering pipeline (fine-tune detectors on new golden sets, run sliding-window inference, apply ITU CPP P2P LOS filter inside a 1 km radius, bin schools by the resulting binary flag). No equation or claim reduces by construction to its own inputs. School and tower mAP@50 figures (0.913 / 0.971) are measured on held-out Lesotho test tiles that were never used to train the base models. The 9/11 high-potential / hard-to-connect split is simply the output of the authors' own definition of the proxy; they do not present it as an external prediction of ground-truth connectivity. The only self-citation of note is the reuse of the USA-trained YOLO26n weights from the authors' concurrent arXiv [7] (and the associated GitHub [30]) as the initialization for fine-tuning; that citation supplies starting parameters, not the Lesotho labels or the final triage numbers. Hyper-parameter citations [31,32] are likewise co-author work but are used only for optimization, not as uniqueness theorems. No fitted scalar is renamed a prediction, no uniqueness result is imported to forbid alternatives, and no known empirical pattern is merely re-labeled. Score 2 therefore reflects a single non-load-bearing self-citation of base weights; the central empirical claims stand independently.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

The central claim rests on three operational free parameters (threshold, radius, stride), a small manually curated golden set, and the domain assumption that LOS-to-tower is a usable connectivity proxy. No new physical entities are postulated; the work is an engineering pipeline built on standard detectors and an existing ITU tool.

free parameters (4)
  • detection confidence threshold = 0.6
    Fixed at 0.6 for both school and tower inference to favor high recall; directly controls which objects enter the connectivity stage.
  • connectivity search radius = 1 km
    1 km zone around each school inside which towers are considered; chosen by hand and stated as adjustable.
  • sliding-window stride = 256 pixels
    256 px on 512 px tiles; determines coverage and duplicate rate for school detection.
  • Lesotho golden-set sizes = 200 train / 93 val / 129 test (schools); 179 tiles (towers)
    200/93/129 school images and 179 tower tiles manually curated; all reported mAP numbers are conditioned on this small labeled pool.
axioms (4)
  • domain assumption 0.6 m satellite imagery contains sufficient visual features for reliable school and mast-tower detection after transfer learning.
    Foundational premise of the vision-only claim; invoked throughout §III–IV.
  • ad hoc to paper Confirmed line-of-sight to a cell tower within 1 km (via CPP P2P + DEM) is an adequate proxy for connectivity potential.
    Defines the high-potential / hard-to-connect labels in §IV.C; discussion (§VI) treats it as triage only.
  • domain assumption Fine-tuning a USA-trained school detector and a Mozambique-trained tower detector with a few hundred local labels yields models that generalize inside Lesotho.
    Justified by Table I ablation but still an empirical assumption for any new geography.
  • standard math Standard YOLO / OBB training losses and non-max-suppression produce usable centroids for subsequent distance and LOS calculations.
    Common computer-vision practice; no novel derivation required.

reviewed 2026-07-12 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Signal from Space: Detecting Schools and Towers to Bridge the Digital Divide." pith.science (2026). https://pith.science/paper/RVAVEQ5M

@misc{pith2026260702724,
  author       = {Pith},
  title        = {Pith review of: Signal from Space: Detecting Schools and Towers to Bridge the Digital Divide},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RVAVEQ5M}},
  note         = {Machine review of arXiv:2607.02724}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Reliable internet access is essential for modern education, yet millions of school-aged children especially in developing regions remain offline due to unconnected schools. The Giga Initiative aims to connect every school to the internet, but doing so at scale requires efficient methods to map schools and assess surrounding connectivity infrastructure without relying on sparse or noisy third-party datasets. In this work, we propose a scalable, vision-only framework that uses high-resolution satellite imagery and transfer learning to address both tasks simultaneously. By adapting pre-trained object detection models to new geographical regions with minimal labeled data, we detect schools and cell towers directly from space. We then analyze the spatial relationship between detected schools and nearby towers as a proxy for connectivity availability. This purely imagery-driven pipeline enables large-scale infrastructure mapping, reduces dependency on auxiliary data, and supports data-driven prioritization of connectivity investments in underserved areas. Our approach is demonstrated on real satellite imagery from Lesotho, showing strong performance across this region.

Figures

Figures reproduced from arXiv: 2607.02724 by Fares Fourati, Mohamed-Slim Alouini, Sandor Farkas, Vladimir Daigele, Walid Mathlouthi, Zakarya Elmimouni.

Figure 1
Figure 1. Figure 1: Inference Pipeline for School Detection and Connectivity Assessment. The process consists of three main stages: (1) Detection of schools over a large area using sliding search windows on satellite imagery [7];(2) Tower search, line-of-sight visibility analysis, and ground distance calculation within a 1km connectivity zone to identify the nearest visible cell tower; and (3) generation of a priority map and… view at source ↗
Figure 2
Figure 2. Figure 2: Spatial Extent of the Research Area in Mafeteng District, Lesotho a) Stage 1: School detection over the entire region: We divide the region of interest into overlapping tiles of fixed size ( 512 × 512 pixels at 0.6 m resolution). A sliding window stride of 256 pixels is used to ensure coverage of schools that may lie on tile boundaries. Each tile is passed through our school detector. For every detected sc… view at source ↗
Figure 3
Figure 3. Figure 3: Geographic Distribution of Schools and Cell Towers with Confirmed Line-of-Sight, Highlighting Likely Connectable and Hard-to￾Connect Schools. b) Stage 2: Cell tower detection and Connectivity assess￾ment: For each detected school, we detect cell towers within a 1 km radius around the school. Rather than relying on simple geodesic distance between school and tower coordinates, we use the ITU CPP P2P radio v… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

35 extracted references · 7 linked inside Pith

  1. [1]

    Measuring digital development: Facts and figures 2025, 2025

    International Telecommunication Union (ITU). Measuring digital development: Facts and figures 2025, 2025. URL https://www.itu.int/itu-d/reports/statistics/facts-figures-2 025/. Accessed: 2026-05-29

    2025

  2. [2]

    Covid-19: Are children able to continue learn- ing during school closures?, 2020

    UNICEF. Covid-19: Are children able to continue learn- ing during school closures?, 2020. URL https://data.uni cef.org/resources/remote-learning-reachability-factsheet/. UNICEF Data Brief

    2020

  3. [3]

    Two thirds of the world’s school-age children have no internet access at home, new unicef-itu report says, 2020

    UNICEF and ITU. Two thirds of the world’s school-age children have no internet access at home, new unicef-itu report says, 2020. URL https://www.unicef.org/turkiye /en/press-releases/two-thirds-worlds-school-age-childre n-have-no-internet-access-home-new-unicef-itu

    2020

  4. [4]

    Giga interactive map

    UNICEF and ITU. Giga interactive map. https://maps.g iga.global/map, 2024. Accessed: 2025-07-28

    2024

  5. [5]

    Bridging the digital divide: A data-driven approach to transformational connectivity, 10 2025

    International Telecommunication Union (ITU) and Er- icsson. Bridging the digital divide: A data-driven approach to transformational connectivity, 10 2025. URL https://www.itu.int/itu-d/sites/cpp/wp-content/uploads /sites/61/2025/10/Bridging-the-digital-divide-CPP.pdf. Connectivity Planning Platform (CPP) Brief

    2025

  6. [6]

    Ai-powered school mapping and connectivity status prediction using earth observation

    Kelsey Doerksen, Isabelle Tingzon, Casper Fibaek, Rochelle Schneider, and Do-Hyung Kim. Ai-powered school mapping and connectivity status prediction using earth observation. InICLR 2024 Workshop on Machine Learning for Remote Sensing (ML4RS). UNICEF and ESA, 2024

    2024

  7. [7]

    Label-efficient school detection from aerial imagery via weakly supervised pretraining and fine-tuning,

    Zakarya Elmimouni, Fares Fourati, and Mohamed-Slim Alouini. Label-efficient school detection from aerial imagery via weakly supervised pretraining and fine-tuning,

  8. [8]

    URL https://arxiv.org/abs/2605.03968

    Pith/arXiv arXiv

  9. [9]

    Connec- tivity planning platform user guide, 2025

    International Telecommunication Union (ITU). Connec- tivity planning platform user guide, 2025. URL https://fn s-division.github.io/cpp-user-guide/#the-platform-model. Accessed: 2026-05-29

    2025

  10. [10]

    A comprehensive review of object detection with traditional 6 and deep learning methods.Signal Processing, 237: 110075, 2025

    Vrushali Pagire, Murthy Chavali, and Ashish Kale. A comprehensive review of object detection with traditional 6 and deep learning methods.Signal Processing, 237: 110075, 2025. ISSN 0165-1684. doi: https://doi.or g/10.1016/j.sigpro.2025.110075. URL https://www.scienc edirect.com/science/article/pii/S0165168425001896

    doi:10.1016/j.sigpro.2025.110075 2025

  11. [11]

    Agricultural object detection with you only look once (yolo) algorithm: A bibliometric and systematic literature review.Computers and Electronics in Agriculture, 223: 109090, 2024

    Chetan M Badgujar, Alwin Poulose, and Hao Gan. Agricultural object detection with you only look once (yolo) algorithm: A bibliometric and systematic literature review.Computers and Electronics in Agriculture, 223: 109090, 2024

    2024

  12. [12]

    A comprehensive systematic review of yolo for medical object detection (2018 to 2023).IEEE Access, 12:57815–57836, 2024

    Mohammed Gamal Ragab, Said Jadid Abdulkadir, Amgad Muneer, Alawi Alqushaibi, Ebrahim Hamid Sumiea, Rizwan Qureshi, Safwan Mahmood Al-Selwi, and Hitham Alhussian. A comprehensive systematic review of yolo for medical object detection (2018 to 2023).IEEE Access, 12:57815–57836, 2024

    2018

  13. [13]

    Scnet: A lightweight and efficient object detection network for remote sensing

    Shiliang Zhu and Ming Miao. Scnet: A lightweight and efficient object detection network for remote sensing. IEEE Geoscience and Remote Sensing Letters, 21:1–5,

  14. [14]

    URL https: //doi.org/10.1109/LGRS.2023.3344937

    doi: 10.1109/LGRS.2023.3344937. URL https: //doi.org/10.1109/LGRS.2023.3344937

    doi:10.1109/lgrs.2023.3344937 2023

  15. [15]

    Yolo-rs: Remote sensing enhanced crop detection methods, 2025

    Linlin Xiao, Zhang Tiancong, Yutong Jia, Xinyu Nie, Mengyao Wang, and Xiaohang Shao. Yolo-rs: Remote sensing enhanced crop detection methods, 2025. URL https://arxiv.org/abs/2504.11165

    Pith/arXiv arXiv 2025

  16. [16]

    Continental-scale building detection from high resolution satellite imagery.arXiv preprint arXiv:2107.12283, 2021

    Wojciech Sirko, Sergii Kashubin, Marvin Ritter, Abigail Annkah, Yasser Salah Eddine Bouchareb, Yann Dauphin, Daniel Keysers, Maxim Neumann, Moustapha Cisse, and John Quinn. Continental-scale building detection from high resolution satellite imagery.arXiv preprint arXiv:2107.12283, 2021

    Pith/arXiv arXiv 2021

  17. [17]

    Grounding dino: Marrying dino with grounded pre-training for open-set object detection

    Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Qing Jiang, Chunyuan Li, Jianwei Yang, Hang Su, et al. Grounding dino: Marrying dino with grounded pre-training for open-set object detection. In European conference on computer vision, pages 38–55. Springer, 2024

    2024

  18. [18]

    End-to-end object detection with transformers, 2020

    Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nico- las Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers, 2020. URL https://arxiv.org/abs/2005.12872

    Pith/arXiv arXiv 2020

  19. [19]

    Yolov10: Real-time end-to-end object detection.Advances in Neural Information Processing Systems, 37:107984–108011, 2024

    Ao Wang, Hui Chen, Lihao Liu, Kai Chen, Zijia Lin, Jungong Han, et al. Yolov10: Real-time end-to-end object detection.Advances in Neural Information Processing Systems, 37:107984–108011, 2024

    2024

  20. [20]

    Ultralytics yolo11, 2024

    Glenn Jocher and Jing Qiu. Ultralytics yolo11, 2024. URL https://github.com/ultralytics/ultralytics

    2024

  21. [21]

    Ultralytics yolo12: Attention-centric real-time object detectors, 2025

    Yunjie Tian, Qixiang Ye, and David Doermann. Ultralytics yolo12: Attention-centric real-time object detectors, 2025. URL https://docs.ultralytics.com/models/yolo12/

    2025

  22. [22]

    Rich feature hierarchies for accurate object detection and semantic segmentation

    Ross Girshick, Jeff Donahue, Trevor Darrell, and Jitendra Malik. Rich feature hierarchies for accurate object detection and semantic segmentation. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 580–587, 2014

    2014

  23. [23]

    Fast r-cnn, 2015

    Ross Girshick. Fast r-cnn, 2015. URL https://arxiv.org/ abs/1504.08083

    Pith/arXiv arXiv 2015

  24. [24]

    Faster r-cnn: Towards real-time object detection with region proposal networks, 2016

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object detection with region proposal networks, 2016. URL https://arxiv. org/abs/1506.01497

    Pith/arXiv arXiv 2016

  25. [25]

    Nguyen, Jessica Block, Daniel Crawl, Naroa Zurutuza, Dohyung Kim, Gordon Han- son, and Ilkay Altintas

    Mehrdad Yazdani, Mai H. Nguyen, Jessica Block, Daniel Crawl, Naroa Zurutuza, Dohyung Kim, Gordon Han- son, and Ilkay Altintas. Scalable detection of rural schools in africa using convolutional neural networks and satellite imagery. In2018 IEEE/ACM International Conference on Utility and Cloud Computing Compan- ion (UCC Companion), pages 1–6. IEEE, 2018. d...

    doi:10.1109/ucc-companion.2018.00048 2018

  26. [26]

    Towards equitable access to information and opportunity for all: map- ping schools with high-resolution satellite imagery and machine learning

    Zhuangfang Yi, Naroa Zurutuza, Drew Bollinger, Manuel Garcia-Herranz, and Dohyung Kim. Towards equitable access to information and opportunity for all: map- ping schools with high-resolution satellite imagery and machine learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, pages 60–66, 2019

    2019

  27. [27]

    Auto- mated school location mapping at scale from satellite imagery based on deep learning.Remote Sensing, 14 (4):897, 2022

    Iyke Maduako, Zhuangfang Yi, Naroa Zurutuza, Shilpa Arora, Christopher Fabian, and Do-Hyung Kim. Auto- mated school location mapping at scale from satellite imagery based on deep learning.Remote Sensing, 14 (4):897, 2022. doi: 10.3390/rs14040897. URL https://www.mdpi.com/2072-4292/14/4/897

    doi:10.3390/rs14040897 2022

  28. [28]

    Large- scale school mapping using weakly supervised deep learn- ing for universal school connectivity

    Isabelle Tingzon, Utku Can Ozturk, and Ivan Dotu. Large- scale school mapping using weakly supervised deep learn- ing for universal school connectivity. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 28449–28457, 2025

    2025

  29. [29]

    Towards interpretable and reliable deep learning models

    Suhaib Amin, Shaker El-Sappagh, Shariful Islam, et al. Towards interpretable and reliable deep learning models. Frontiers in Neuroscience, 17:1293803, 2023. doi: 10.3 389/fnins.2023.1293803. URL https://pubmed.ncbi.nlm. nih.gov/38077598/

    arXiv 2023

  30. [30]

    Enabling decision-support systems through automated cell tower detection.arXiv preprint arXiv:2311.07840, 2023

    Natasha Krell, Will Gleave, Daniel Nakada, Justin Downes, Amanda Willet, and Matthew Baran. Enabling decision-support systems through automated cell tower detection.arXiv preprint arXiv:2311.07840, 2023. URL https://arxiv.org/abs/2311.07840

    Pith/arXiv arXiv 2023

  31. [31]

    Giga interactive map

    GIGA. Giga interactive map. https://maps.giga.global/d ocs/explore-api, 2024. Accessed: 2025-07-28

    2024

  32. [32]

    School detection: Code and re- sources for school detection from aerial imagery

    Zakarya Elmimouni. School detection: Code and re- sources for school detection from aerial imagery. https: //github.com/zakarya-elmimouni/School_Detection, 2026. GitHub repository, accessed June 25, 2026

    2026

  33. [33]

    Every call is precious: Global optimization of black-box functions with unknown lips- chitz constants

    Fares Fourati, Salma Kharrat, Vaneet Aggarwal, and Mohamed-Slim Alouini. Every call is precious: Global optimization of black-box functions with unknown lips- chitz constants. InInternational Conference on Artificial Intelligence and Statistics, pages 5176–5184. PMLR, 2025

    2025

  34. [34]

    Ecpv2: Fast, efficient, and scalable global optimization of lipschitz functions.Proceedings of the AAAI Conference on Artificial Intelligence, 40(43):36909– 36918, Mar

    Fares Fourati, Mohamed-Slim Alouini, and Vaneet Ag- garwal. Ecpv2: Fast, efficient, and scalable global optimization of lipschitz functions.Proceedings of the AAAI Conference on Artificial Intelligence, 40(43):36909– 36918, Mar. 2026. doi: 10.1609/aaai.v40i43.41018. URL https://ojs.aaai.org/index.php/AAAI/article/view/41018

    doi:10.1609/aaai.v40i43.41018 2026

  35. [35]

    Oriented bounding boxes (obb) object detec- tion

    Ultralytics. Oriented bounding boxes (obb) object detec- tion. https://docs.ultralytics.com/tasks/obb, 2026

    2026

This paper was first reviewed by grok-4.5 on July 12, 2026.