arxiv: 2605.12276 · v1 · submitted 2026-05-12 · 💻 cs.AI

Recognition: 2 theorem links

· Lean Theorem

NARA: Anchor-Conditioned Relation-Aware Contextualization of Heterogeneous Geoentities

Jina Kim , Gengchen Mai , Lingyi Zhao , Khurram Shafique , Yao-Yi Chiang

Authors on Pith no claims yet

Pith reviewed 2026-05-13 04:42 UTC · model grok-4.3

classification 💻 cs.AI

keywords vector geospatial dataself-supervised learningspatial relationsrepresentation learningheterogeneous geoentitiescontext-dependent representationsgeospatial foundation models

0 comments

The pith

NARA learns context-dependent representations for geospatial entities by jointly modeling semantics, geometry, and spatial relations in a unified self-supervised framework.

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

The paper proposes NARA as a self-supervised framework to overcome fragmented existing methods for vector geospatial data. It claims that jointly modeling semantics, geometry, and spatial relations allows the capture of relational structures beyond mere proximity, producing richer contextual representations usable across points, polylines, and polygons. A sympathetic reader would care because this unified approach could improve accuracy on real tasks involving geographic entities. The work shows consistent gains on building function classification, traffic speed prediction, and next point-of-interest recommendation.

Core claim

NARA (Neural Anchor-conditioned Relation-Aware representation learning) is a self-supervised framework for vector geoentities. It learns context-dependent representations by jointly modeling semantics, geometry, and spatial relations within a unified framework and captures relational spatial structure beyond proximity alone, enabling rich contextualized representations across heterogeneous geoentities of points, polylines, and polygons. Evaluation on building function classification, traffic speed prediction, and next point-of-interest recommendation shows consistent improvements over prior methods.

What carries the argument

The NARA framework, which uses anchor-conditioned relation-aware modeling to integrate semantics, geometry, and spatial relations for self-supervised learning on heterogeneous vector geoentities.

If this is right

Consistent performance gains on building function classification tasks.
Improved accuracy in traffic speed prediction applications.
Better results for next point-of-interest recommendation systems.
Unified handling of points, polylines, and polygons without separate models.

Where Pith is reading between the lines

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

The framework could serve as a basis for larger-scale geospatial foundation models focused on vector data.
Extensions might test additional topological relations or integration with raster data sources.
The joint modeling strategy could apply to other domains with mixed entity types and structured relations.

Load-bearing premise

That jointly modeling semantics, geometry, and spatial relations in a self-supervised manner will produce representations that generalize better than fragmented prior methods on downstream tasks.

What would settle it

If NARA fails to show consistent improvements over prior methods on building function classification, traffic speed prediction, or next point-of-interest recommendation, or if it does not effectively capture relational spatial structures beyond proximity.

Figures

Figures reproduced from arXiv: 2605.12276 by Gengchen Mai, Jina Kim, Khurram Shafique, Lingyi Zhao, Yao-Yi Chiang.

read the original abstract

Geospatial foundation models have primarily focused on raster data such as satellite imagery, where self-supervised learning has been widely studied. Vector geospatial data instead represent the world as discrete geoentities with explicit geometry, semantics, and structured spatial relations, including metric proximity and topological relationships. These relations jointly determine how entities interact within space, yet existing representation learning methods remain fragmented, often restricted to specific geometry types or partial spatial relations, limiting their ability to capture unified spatial context across heterogeneous geoentities. We propose NARA (Neural Anchor-conditioned Relation-Aware representation learning), a self-supervised framework for vector geoentities. NARA learns context-dependent representations by jointly modeling semantics, geometry, and spatial relations within a unified framework and captures relational spatial structure beyond proximity alone, enabling rich contextualized representations across heterogeneous geoentities of points, polylines, and polygons. Evaluation on building function classification, traffic speed prediction, and next point-of-interest recommendation shows consistent improvements over prior methods, highlighting the benefit of unified relational modeling for vector geospatial data.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

NARA unifies self-supervised modeling of semantics, geometry, and spatial relations across points, lines, and polygons, but the abstract leaves the actual gains and implementation details thin.

read the letter

The core contribution is a single self-supervised framework that conditions representations on anchors while capturing both metric and topological relations for mixed geometry types. This directly tackles the fragmentation the abstract describes, where prior methods typically handle only one geometry class or rely mainly on proximity. If the full paper shows how the joint modeling actually works without separate pipelines, that is a useful step for vector geospatial work in urban applications like traffic or POI tasks.

Referee Report

1 major / 1 minor

Summary. The manuscript proposes NARA, a self-supervised framework for vector geospatial data that learns context-dependent representations of heterogeneous geoentities (points, polylines, polygons) by jointly modeling semantics, geometry, and spatial relations including topological relationships beyond proximity. It evaluates the approach on three downstream tasks—building function classification, traffic speed prediction, and next point-of-interest recommendation—claiming consistent improvements over prior fragmented methods.

Significance. If the quantitative results and architectural details hold up under scrutiny, the work would address a genuine gap in geospatial foundation models by shifting focus from raster data to structured vector representations with explicit relational modeling. This unified treatment of semantics, geometry, and topology could improve generalization on tasks requiring spatial context, though the abstract provides no metrics to gauge the effect size.

major comments (1)

[Abstract] Abstract: the central claim of 'consistent improvements over prior methods' is asserted without any quantitative results, baselines, error bars, statistical tests, or even high-level architectural/loss details. This absence makes it impossible to assess whether the data support the claim that joint modeling of semantics, geometry, and spatial relations yields richer representations.

minor comments (1)

[Abstract] The abstract and title use dense terminology ('anchor-conditioned relation-aware contextualization') without a concise one-sentence definition of the core mechanism; a short illustrative example of how an anchor conditions relations for a polyline would improve accessibility.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive comments. We agree that the abstract would be strengthened by including quantitative highlights and will revise it accordingly to better support our claims while maintaining conciseness.

read point-by-point responses

Referee: [Abstract] Abstract: the central claim of 'consistent improvements over prior methods' is asserted without any quantitative results, baselines, error bars, statistical tests, or even high-level architectural/loss details. This absence makes it impossible to assess whether the data support the claim that joint modeling of semantics, geometry, and spatial relations yields richer representations.

Authors: We agree that the abstract should provide quantitative context to support the claim of consistent improvements. In the revised version, we will incorporate specific performance gains (e.g., relative improvements on building classification, traffic prediction, and POI recommendation), reference the main baselines, and note that results include standard error reporting from multiple runs. We will also add a concise high-level description of the anchor-conditioned relation-aware objective and the joint modeling of semantics/geometry/topology. These additions will be kept brief to respect abstract length limits while enabling readers to gauge effect sizes. The full experimental details, tables with error bars, and statistical comparisons remain in the main body and appendix. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper proposes NARA, a new self-supervised framework for learning context-dependent representations of heterogeneous vector geoentities by jointly modeling semantics, geometry, and spatial relations (including topology beyond proximity). The central claims rest on the architecture description and empirical gains on three downstream tasks (building function classification, traffic speed prediction, next POI recommendation) rather than any derivation that reduces to fitted parameters, self-definitions, or self-citation chains. No load-bearing step equates a prediction to its input by construction, and the work is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no specific details on model components, loss functions, or training assumptions, so the ledger cannot be populated with concrete entries.

pith-pipeline@v0.9.0 · 5488 in / 1077 out tokens · 61230 ms · 2026-05-13T04:42:44.702241+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

NARA learns context-dependent representations by jointly modeling semantics, geometry, and spatial relations within a unified framework and captures relational spatial structure beyond proximity alone
IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

semivariogram regularization... sibling groups Sa,r defined by topological relation rel(vi,a)

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

48 extracted references · 48 canonical work pages · 1 internal anchor

[1]

Planet dump from openstreetmap.https://planet.osm.org,

OpenStreetMap contributors. Planet dump from openstreetmap.https://planet.osm.org,

work page
[2]

Accessed: 2026-03-26

work page 2026
[3]

Overture maps open data platform

OpenStreetMap contributors, Overture Maps Foundation. Overture maps open data platform. https://overturemaps.org, 2026. Accessed: 2026-03-26

work page 2026
[4]

John Wiley & Sons, 2015

Paul A Longley, Michael F Goodchild, David J Maguire, and David W Rhind.Geographic information science and systems. John Wiley & Sons, 2015

work page 2015
[5]

Rethinking transformers pre-training for multi-spectral satellite imagery

Mubashir Noman, Muzammal Naseer, Hisham Cholakkal, Rao Muhammad Anwer, Salman Khan, and Fahad Shahbaz Khan. Rethinking transformers pre-training for multi-spectral satellite imagery. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 27811–27819, 2024

work page 2024
[6]

Csp: Self-supervised contrastive spatial pre-training for geospatial-visual representations

Gengchen Mai, Ni Lao, Yutong He, Jiaming Song, and Stefano Ermon. Csp: Self-supervised contrastive spatial pre-training for geospatial-visual representations. InInternational Conference on Machine Learning, pages 23498–23515. PMLR, 2023

work page 2023
[7]

Gair: Location-aware self-supervised contrastive pre-training with geo-aligned implicit representations.ISPRS Journal of Photogrammetry and Remote Sensing, 2026

Zeping Liu, Lao Ni, Zhangyu Wang, Junfeng Jiao, and Gengchen Mai. Gair: Location-aware self-supervised contrastive pre-training with geo-aligned implicit representations.ISPRS Journal of Photogrammetry and Remote Sensing, 2026

work page 2026
[8]

Satclip: Global, general-purpose location embeddings with satellite imagery

Konstantin Klemmer, Esther Rolf, Caleb Robinson, Lester Mackey, and Marc Rußwurm. Satclip: Global, general-purpose location embeddings with satellite imagery. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 4347–4355, 2025

work page 2025
[9]

Prithvi-eo-2.0: A versatile multi-temporal foundation model for earth observation applications.IEEE Transactions on Geoscience and Remote Sensing, 2025

Daniela Szwarcman, Sujit Roy, Paolo Fraccaro, Orsteinn Elí Gíslason, Benedikt Blumenstiel, Rinki Ghosal, Pedro Henrique De Oliveira, Joao Lucas de Sousa Almeida, Rocco Sedona, Yanghui Kang, et al. Prithvi-eo-2.0: A versatile multi-temporal foundation model for earth observation applications.IEEE Transactions on Geoscience and Remote Sensing, 2025

work page 2025
[10]

BERT: Pre-training of deep bidirectional transformers for language understanding

Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep bidirectional transformers for language understanding. In Jill Burstein, Christy Doran, and Thamar Solorio, editors,Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volu...

work page 2019
[11]

An image is worth 16x16 words: Transformers for image recognition at scale

Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale. InInternational Conference on Learning Representations, 2021

work page 2021
[12]

Multi-scale repre- sentation learning for spatial feature distributions using grid cells

Gengchen Mai, Krzysztof Janowicz, Bo Yan, Rui Zhu, Ling Cai, and Ni Lao. Multi-scale repre- sentation learning for spatial feature distributions using grid cells. InInternational Conference on Learning Representations, 2020

work page 2020
[13]

SpaBERT: A pretrained language model from geographic data for geo-entity representation

Zekun Li, Jina Kim, Yao-Yi Chiang, and Muhao Chen. SpaBERT: A pretrained language model from geographic data for geo-entity representation. In Yoav Goldberg, Zornitsa Kozareva, and Yue Zhang, editors,Findings of the Association for Computational Linguistics: EMNLP 2022, pages 2757–2769, Abu Dhabi, United Arab Emirates, December 2022. Association for Compu...

work page 2022
[14]

Poi-enhancer: An llm-based semantic enhancement framework for poi represen- tation learning.Proceedings of the AAAI Conference on Artificial Intelligence, 39(11):11509– 11517, Apr

Jiawei Cheng, Jingyuan Wang, Yichuan Zhang, Jiahao Ji, Yuanshao Zhu, Zhibo Zhang, and Xiangyu Zhao. Poi-enhancer: An llm-based semantic enhancement framework for poi represen- tation learning.Proceedings of the AAAI Conference on Artificial Intelligence, 39(11):11509– 11517, Apr. 2025

work page 2025
[15]

Road network representation learning: A dual graph-based approach.ACM Transactions on Knowledge Discovery from Data, 17(9):1–25, 2023

Liang Zhang and Cheng Long. Road network representation learning: A dual graph-based approach.ACM Transactions on Knowledge Discovery from Data, 17(9):1–25, 2023. 10

work page 2023
[16]

Road network representation learning with the third law of geography

Haicang Zhou, Weiming Huang, Yile Chen, Tiantian He, Gao Cong, and Yew-Soon Ong. Road network representation learning with the third law of geography. In A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang, editors,Advances in Neural Information Processing Systems, volume 37, pages 11789–11813. Curran Associates, Inc., 2024

work page 2024
[17]

Urban region representation learning with openstreetmap building footprints

Yi Li, Weiming Huang, Gao Cong, Hao Wang, and Zheng Wang. Urban region representation learning with openstreetmap building footprints. InProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD ’23, page 1363–1373, New York, NY , USA, 2023. Association for Computing Machinery

work page 2023
[18]

Geovectors: A linked open corpus of openstreetmap embeddings on world scale

Nicolas Tempelmeier, Simon Gottschalk, and Elena Demidova. Geovectors: A linked open corpus of openstreetmap embeddings on world scale. InProceedings of the 30th ACM Inter- national Conference on Information & Knowledge Management, CIKM ’21, page 4604–4612, New York, NY , USA, 2021. Association for Computing Machinery

work page 2021
[19]

City foundation models for learning general purpose representations from openstreetmap

Pasquale Balsebre, Weiming Huang, Gao Cong, and Yi Li. City foundation models for learning general purpose representations from openstreetmap. InProceedings of the 33rd ACM Interna- tional Conference on Information and Knowledge Management, CIKM ’24, page 87–97, New York, NY , USA, 2024. Association for Computing Machinery

work page 2024
[20]

Hygmap: representing all types of map entities via heterogeneous hypergraph

Yifan Yang, Jingyuan Wang, Xie Yu, and Yibang Tang. Hygmap: representing all types of map entities via heterogeneous hypergraph. InProceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, IJCAI ’25, 2025

work page 2025
[21]

Self-supervised representa- tion learning for geospatial objects: A survey.Information Fusion, 123:103265, 2025

Yile Chen, Weiming Huang, Kaiqi Zhao, Yue Jiang, and Gao Cong. Self-supervised representa- tion learning for geospatial objects: A survey.Information Fusion, 123:103265, 2025

work page 2025
[22]

node2vec: Scalable feature learning for networks

Aditya Grover and Jure Leskovec. node2vec: Scalable feature learning for networks. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ’16, page 855–864, New York, NY , USA, 2016. Association for Computing Machinery

work page 2016
[23]

Learning embeddings of intersections on road networks

Meng-xiang Wang, Wang-Chien Lee, Tao-yang Fu, and Ge Yu. Learning embeddings of intersections on road networks. InProceedings of the 27th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems, SIGSPATIAL ’19, page 309–318, New York, NY , USA, 2019. Association for Computing Machinery

work page 2019
[24]

Jensen, and Thomas Dyhre Nielsen

Tobias Skovgaard Jepsen, Christian S. Jensen, and Thomas Dyhre Nielsen. Relational fusion networks: Graph convolutional networks for road networks.IEEE Transactions on Intelligent Transportation Systems, 23(1):418–429, 2022

work page 2022
[25]

Towards general-purpose representation learning of polygonal geometries.GeoInformatica, 27(2):289–340, 2023

Gengchen Mai, Chiyu Jiang, Weiwei Sun, Rui Zhu, Yao Xuan, Ling Cai, Krzysztof Janowicz, Stefano Ermon, and Ni Lao. Towards general-purpose representation learning of polygonal geometries.GeoInformatica, 27(2):289–340, 2023

work page 2023
[26]

Poly2vec: Polymorphic fourier-based encoding of geospatial objects for geoAI applications

Maria Despoina Siampou, Jialiang Li, John Krumm, Cyrus Shahabi, and Hua Lu. Poly2vec: Polymorphic fourier-based encoding of geospatial objects for geoAI applications. InForty- second International Conference on Machine Learning, 2025

work page 2025
[27]

Geo2vec: Shape-and distance-aware neural representation of geospatial entities

Chen Chu and Cyrus Shahabi. Geo2vec: Shape-and distance-aware neural representation of geospatial entities. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 18985–18993, 2026

work page 2026
[28]

From itdl to place2vec: Reasoning about place type similarity and relatedness by learning embeddings from augmented spatial contexts

Bo Yan, Krzysztof Janowicz, Gengchen Mai, and Song Gao. From itdl to place2vec: Reasoning about place type similarity and relatedness by learning embeddings from augmented spatial contexts. InProceedings of the 25th ACM SIGSPATIAL international conference on advances in geographic information systems, pages 1–10, 2017

work page 2017
[29]

Relaxing unanswerable geographic questions using a spatially explicit knowledge graph embedding model

Gengchen Mai, Bo Yan, Krzysztof Janowicz, and Rui Zhu. Relaxing unanswerable geographic questions using a spatially explicit knowledge graph embedding model. InInternational conference on geographic information science, pages 21–39. Springer, 2019. 11

work page 2019
[30]

Estimating urban functional distributions with semantics preserved poi embedding.International Journal of Geographical Information Science, 36(10):1905–1930, 2022

Weiming Huang, Lizhen Cui, Meng Chen, Daokun Zhang, and Yao Yao. Estimating urban functional distributions with semantics preserved poi embedding.International Journal of Geographical Information Science, 36(10):1905–1930, 2022

work page 1905
[31]

Tile2vec: Unsupervised representation learning for spatially distributed data

Neal Jean, Sherrie Wang, Anshul Samar, George Azzari, David Lobell, and Stefano Ermon. Tile2vec: Unsupervised representation learning for spatially distributed data. InProceedings of the AAAI Conference on Artificial Intelligence, volume 33, pages 3967–3974, 2019

work page 2019
[32]

Mc-gta: Metric-constrained model-based clustering using goodness-of-fit tests with autocorrelations

Zhangyu Wang, Gengchen Mai, Krzysztof Janowicz, and Ni Lao. Mc-gta: Metric-constrained model-based clustering using goodness-of-fit tests with autocorrelations. In41st International Conference on Machine Learning, ICML 2024, 2024

work page 2024
[33]

Rethink geographical generalizability with unsupervised self-attention model ensemble: A case study of openstreetmap missing building detection in africa

Hao Li, Jiapan Wang, Johann Maximilian Zollner, Gengchen Mai, Ni Lao, and Martin Werner. Rethink geographical generalizability with unsupervised self-attention model ensemble: A case study of openstreetmap missing building detection in africa. InProceedings of the 31st ACM International Conference on Advances in Geographic Information Systems, pages 1–9, 2023

work page 2023
[34]

A computer movie simulating urban growth in the detroit region.Economic geography, 46(sup1):234–240, 1970

Waldo R Tobler. A computer movie simulating urban growth in the detroit region.Economic geography, 46(sup1):234–240, 1970

work page 1970
[35]

On the first law of geography: A reply.Annals of the association of American geographers, 94(2):304–310, 2004

Waldo Tobler. On the first law of geography: A reply.Annals of the association of American geographers, 94(2):304–310, 2004

work page 2004
[36]

Variogram or semivariogram? variance or semivari- ance? allan variance or introducing a new term?Mathematical Geosciences, 43(6):735–740, 2011

Martin Bachmaier and Matthias Backes. Variogram or semivariogram? variance or semivari- ance? allan variance or introducing a new term?Mathematical Geosciences, 43(6):735–740, 2011

work page 2011
[37]

A mathematical framework for the definition of topological relations

Max Egenhofer. A mathematical framework for the definition of topological relations. InProc. the fourth international symposium on spatial data handing, pages 803–813, 1990

work page 1990
[38]

A spatial logic based on regions and connection.KR, 92(165-176):40–40, 1992

David A Randell, Zhan Cui, Anthony G Cohn, et al. A spatial logic based on regions and connection.KR, 92(165-176):40–40, 1992

work page 1992
[39]

Computing and querying strict, approximate, and metrically refined topological relations in linked geographic data.Transactions in GIS, 23(3):601–619, 2019

Blake Regalia, Krzysztof Janowicz, and Grant McKenzie. Computing and querying strict, approximate, and metrically refined topological relations in linked geographic data.Transactions in GIS, 23(3):601–619, 2019

work page 2019
[40]

GeoLM: Empowering language models for geospatially grounded language understanding

Zekun Li, Wenxuan Zhou, Yao-Yi Chiang, and Muhao Chen. GeoLM: Empowering language models for geospatially grounded language understanding. In Houda Bouamor, Juan Pino, and Kalika Bali, editors,Proceedings of the 2023 Conference on Empirical Methods in Nat- ural Language Processing, pages 5227–5240, Singapore, December 2023. Association for Computational L...

work page 2023
[41]

Eckel, and José Luis Ambite

Yijun Lin, Yao-Yi Chiang, Meredith Franklin, Sandrah P. Eckel, and José Luis Ambite. Building autocorrelation-aware representations for fine-scale spatiotemporal prediction. In2020 IEEE International Conference on Data Mining (ICDM), pages 352–361, 2020

work page 2020
[42]

Jilin Hu, Chenjuan Guo, Bin Yang, and Christian S. Jensen. Stochastic weight completion for road networks using graph convolutional networks. In2019 IEEE 35th International Conference on Data Engineering (ICDE), pages 1274–1285, 2019

work page 2019
[43]

Modeling user activity preference by leveraging user spatial temporal characteristics in lbsns.IEEE Transactions on Systems, Man, and Cybernetics: Systems, 45(1):129–142, 2014

Dingqi Yang, Daqing Zhang, Vincent W Zheng, and Zhiyong Yu. Modeling user activity preference by leveraging user spatial temporal characteristics in lbsns.IEEE Transactions on Systems, Man, and Cybernetics: Systems, 45(1):129–142, 2014

work page 2014
[44]

Llama 2: Open Foundation and Fine-Tuned Chat Models

Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. Llama 2: Open foundation and fine-tuned chat models.arXiv preprint arXiv:2307.09288, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023
[45]

Pre-training time-aware location embeddings from spatial-temporal trajectories.IEEE Transactions on Knowledge and Data Engineering, 34(11):5510–5523, 2022

Huaiyu Wan, Yan Lin, Shengnan Guo, and Youfang Lin. Pre-training time-aware location embeddings from spatial-temporal trajectories.IEEE Transactions on Knowledge and Data Engineering, 34(11):5510–5523, 2022. 12

work page 2022
[46]

Yan Lin, Huaiyu Wan, Shengnan Guo, and Youfang Lin. Pre-training context and time aware location embeddings from spatial-temporal trajectories for user next location prediction.Pro- ceedings of the AAAI Conference on Artificial Intelligence, 35(5):4241–4248, May 2021

work page 2021
[47]

Mordechai Haklay. How good is volunteered geographical information? a comparative study of openstreetmap and ordnance survey datasets.Environment and planning B: Planning and design, 37(4):682–703, 2010

work page 2010
[48]

Con- struction

Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. InInternational Conference on Learning Representations, 2019. 13 A Limitations & Broader Impact LimitationsNARA is pretrained on a specific geographic region; as with most geospatial mod- els [20], spatial variability across cities with different urban structures may limit transferab...

work page 2019