Pith. sign in

REVIEW 4 major objections 5 minor 94 references

UrbanMind: Towards Urban General Intelligence via Tool-Enhanced Retrieval-Augmented Generation and Multilevel Optimization

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read UrbanMind argues that urban general intelligence can be built by making retrieval, generation, and domain weighting a single nested optimization, with a knowledge base that updates as the city changes.

desk verdict Plausible framework paper with no quantitative support and a malformed multilevel formulation; worth refereeing but not publishing as-is. read the letter →

arxiv 2507.04706 v1 pith:ZQAENLDU submitted 2025-07-07 cs.LG cs.AI

Kai Yang , Zelin Zhu , Chengtao Jian , Hui Ma , Shengjie Zhao , Xiaozhou Ye , Ye Ouyang This is my paper
classification cs.LGcs.AI
keywords urbangeneralintelligenceretrieval-augmentedgenerationcontinuallearningmultileveloptimizationmixture-of-expertstool-enhancedRAGdistributionallyrobustsmartcities
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

UrbanMind is the paper's attempt to make AI systems work in cities that change over time. The paper claims that a retrieval-augmented generation framework, with a continually updated knowledge base and tool calls, can support 'urban general intelligence' — the ability to perceive, reason, and act across dynamic urban tasks. It frames the training as a multilevel optimization problem in which the retriever, the generator, and the weights over data domains are chosen in a nested way, mirroring Mixture-of-Experts LLM training. On three levels of urban tasks and a travel-planning demo, the paper reports that continual RAG with tools responds better than a plain LLM or static RAG.

What carries the argument

The load-bearing object is Eq. (6), a three-level optimization problem. At the top, retriever parameters $\boldsymbol{\theta}$ are chosen to minimize validation loss. In the middle, generator parameters $\boldsymbol{\phi}$ minimize a training loss weighted over $M$ domains. At the bottom, domain weights $\boldsymbol{w}$ are learned under a $\mathrm{KL}(\boldsymbol{w}\|\boldsymbol{p}_{\mathrm{uniform}}) < \epsilon$ constraint to keep the solution robust to distribution shift. This nesting is what makes the framework 'multilevel': each level's optimum is a constraint on the level above, so retrieval, generation, and domain reweighting are tuned together. The machinery also includes an incremental corpus updater that prunes stale entries using temporal decay and redundancy detection, and a tool set that the retriever can invoke for real-time information.

What would settle it

Compute the exact or approximate solution to Eq. (6) on a standard urban benchmark and compare it, under identical corpora and tools, against a single-level retriever and generator trained jointly and against fixed domain weights; if the nested solution does not improve validation loss, retrieval recall, or resistance to distribution shift, the paper's core claim is falsified.

Watch

Extended reading notes

Core claim

The central claim is that urban general intelligence can be approached by a system that never stops learning from its environment. UrbanMind combines three mechanisms: a task-aware retriever that searches a dynamically updated knowledge base, a generator that fuses retrieved knowledge with the query, and a continual adaptation layer that updates model parameters without forgetting old tasks. The paper's distinctive assertion is that these should be optimized as a three-level problem — the retriever's parameters are chosen to minimize validation loss, the generator's parameters are trained under those retrieval choices and domain weights, and the domain weights themselves are learned to be robust to distribution shift under a KL constraint. The paper argues this formulation aligns naturally with MoE LLMs, where a gating network routes inputs to experts, and claims the framework outperforms LLM-only and static-RAG baselines on real-world urban tasks of increasing complexity.

Load-bearing premise

The assumption that the three-level nested optimization in Eq. (6) is tractable and beneficial is load-bearing; the paper offers no algorithm, convergence analysis, or experiment that actually runs this optimization, so the architecture's central advantage is asserted rather than demonstrated.

Editorial extensions

If this is right

  • Urban AI services could be updated by refreshing the knowledge base and re-tuning the retriever, rather than retraining the whole model, making long-term deployment cheaper.
  • The same formulation supports partial optimization: if the generator is fixed, the retriever can still be trained against robust domain weights, and vice versa, so teams with limited compute can still improve part of the system.
  • Cloud-edge deployment becomes a natural fit, with local adapters holding region-specific knowledge and the cloud maintaining global knowledge for the LLM.
  • Tool-enhanced retrieval lets the LLM ground its decisions in real-time conditions — time, weather, traffic — so its answers are actionable rather than generic.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An implication the authors leave implicit is that the real test of Eq. (6) is whether the nested optimization beats simpler training; the paper's demonstrations do not exercise the nested objective, so the framework's core benefit remains unmeasured.
  • The multilevel formulation could be transferred to other non-stationary domains, such as finance or healthcare, wherever a retrieval corpus and tool calls evolve over time — but only if the intermediate generator–retriever coupling is computationally tractable.
  • A testable extension would be to run Eq. (6) with and without the KL-constrained domain weights and compare worst-case performance on shifted domains, which would isolate the DRO layer's contribution.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes UrbanMind, a tool-enhanced retrieval-augmented generation (RAG) framework intended to support urban general intelligence. It describes a four-layer architecture (database, retrieval, integration, adaptation), a continual RAG design called C-RAG-LLM, cloud-edge deployment, and a multilevel optimization formulation (Eq. (6)) intended to couple retriever parameters θ, generator parameters φ, and domain weights w. The paper claims that the architecture aligns naturally with a multilevel optimization framework and that evaluations on real-world urban tasks verify its effectiveness. The reported evaluation consists of qualitative comparisons of LLM-only, static RAG, continual RAG, and a tool-enhanced travel-planning prototype.

Significance. If the formal and empirical claims were supported, UrbanMind would offer a useful organizing framework for continual RAG in urban settings and a concrete deployment architecture. The paper deserves credit for a clear task taxonomy (Levels 1-3), a sensible architecture diagram, and a working travel-planning prototype that illustrates the potential of tool-enhanced RAG. However, the central formal contribution, Eq. (6), is not a well-defined optimization problem as written, no algorithm or convergence analysis is provided, and Section 4 contains no quantitative results at all. The paper does not ship machine-checked proofs, reproducible code, parameter-free derivations, or falsifiable quantitative predictions, so the claimed 'principled' multilevel optimization and 'superior performance' are not established.

major comments (4)
  1. [§3.5, Eq. (6)] The three-level program in Eq. (6) is not a well-defined multilevel optimization problem as written. The innermost problem defines w*(θ) = argmin_w L_dro(w; θ, φ), and L_dro depends on the generator parameters φ; but φ is the decision variable of the middle level, which is in turn constrained by φ*(θ) = argmin_φ L_gen(φ; θ, w*(θ)). Hence w* is defined in terms of φ*, and φ* in terms of w*, producing a circular fixed-point condition rather than a nested hierarchy of lower-level problems. No existence, uniqueness, convergence, or complexity analysis is provided, and the only algorithmic statement in Section 3.5 is that the problem can be 'decomposed or relaxed into a sequence of tractable sub-problems,' without specifying how. The central formal contribution of the paper is therefore not established.
  2. [§4, §4.1, §4.2] The evaluation does not provide any quantitative evidence for the paper's effectiveness claims. Section 4.2 reports only qualitative screenshots (Figs. 5-7 and 9-11), with no tables, numeric metrics, error bars, ablations, or statistical tests. Section 4.1 states that 'Real-time data updates were simulated using synthetic streams generated from publicly available urban datasets [30],' so the abstract's phrase 'real-world urban tasks' and Section 1.6's claim of 'superior performance compared to baseline approaches' are not supported by the reported experiments. The metrics listed in Section 3.2 (Top-k accuracy, MRR, NDCG, drift-aware metrics) are never reported.
  3. [§3.5, Eq. (6)] The robustness claim is disconnected from the formulation. The uncertainty set U is defined as {P : D_KL(P||P_train) ≤ ρ}, but Eq. (6) never optimizes over distributions P in U; it only constrains the domain weight vector w by KL(w||p_uniform) < ε. The DRO objective L_dro(w; θ, φ) is a weighted empirical loss, not a worst-case loss over U. Thus the statement that this formulation 'ensures robust performance under worst-case scenarios within U' is not justified. Additionally, the text introduces λ as a regularization coefficient, but λ does not appear in Eq. (6).
  4. [§2.1 and §3.5] The paper explicitly allows the multi-timescale update hierarchy to invert by domain (e.g., knowledge base updates at the shortest timescale in traffic prediction, model adaptation at the shortest timescale in personalized dialogue), but Eq. (6) fixes a single hierarchy with θ as the outer variable, φ as the middle variable, and w as the inner variable. No mechanism is described for how the formal program changes under inversion, so the claimed flexibility of the multilevel formulation is not realized.
minor comments (5)
  1. [§1.1, §1.2, §2.3] There are numerous typos and grammatical errors, including 'continual adaption' (Section 1.1), 'effective meanings' (Section 1.2), and 'can broadly categorized' (Section 2.3); the manuscript would benefit from a careful proofreading pass.
  2. [References [39] and [40]] References [39] and [40] appear to duplicate the same ICLR paper, and [39] lacks a publication year; please reconcile the duplicate entries.
  3. [ACM Reference Format] The ACM Reference Format line gives a publication date of 2018 and venue J. ACM, which is inconsistent with the 2025 arXiv submission; this should be corrected.
  4. [§4.1] The experimental setup describes only OpenStreetMap data extracts [30] and does not explain how the Level-1/2/3 example questions (e.g., Highway XXX congestion, zoning regulations) are derived from or grounded in that dataset, so the evaluation is not reproducible.
  5. [§3.4] The MoE bilevel formulation in Eq. (4) is generic and is not connected to the implementation of UrbanMind's C-RAG-LLM; the paper should clarify which components of the proposed system correspond to gating and experts, or remove the MoE framing if it is only illustrative.

Circularity Check

1 steps flagged · score 6.0 of 10

Eq. (6) is self-referential: φ*(θ) is defined through w*(θ), which is itself defined through φ, so the central multilevel-optimization formulation is circular as written.

  1. self definitional [Section 3.5, Eq. (6)]
    "s.t. 𝝓∗(𝜽) = arg min 𝝓 Lgen(𝝓; 𝜽, 𝒘∗(𝜽)) = 𝑀∑︁ 𝑚=1 𝑤∗ 𝑚 ∑︁ (𝑞𝑖,𝑎𝑖)∈D𝑚 ℓtrain(𝑎𝑖,G(𝑞𝑖,R(𝑞𝑖 ; 𝜽); 𝝓)), s.t. 𝒘∗(𝜽) = arg min 𝒘 Ldro(𝒘; 𝜽, 𝝓) = 𝑀∑︁ 𝑚=1 𝑤𝑚 ∑︁ (𝑞𝑖,𝑎𝑖)∈D𝑚 ℓtrain(𝑎𝑖,G(𝑞𝑖,R(𝑞𝑖 ; 𝜽); 𝝓)),"

    The defining equations are mutually recursive. The middle-level solution 𝝓*(𝜽) is defined as argmin over 𝝓 of Lgen(𝝓; 𝜽, 𝒘*(𝜽)), so it depends on 𝒘*(𝜽). The inner-level solution 𝒘*(𝜽) is defined as argmin over 𝒘 of Ldro(𝒘; 𝜽, 𝝓), whose objective contains 𝝓 — the very variable being solved in the middle level. Thus the lower-level solution is not a function of the outer variables only; it depends on the middle-level optimizer, while the middle-level optimizer depends on the lower-level solution. This is a circular fixed-point condition, not a nested multilevel hierarchy. Section 3.5 acknowledges only that the problem 'can be decomposed or relaxed into a sequence of tractable sub-problems' and gives no existence, uniqueness, or algorithm for the general form. Because Eq.

full rationale

The UrbanMind architecture, continual corpus updating, tool-enhanced travel prototype, and qualitative comparisons are largely self-contained: no fitted parameter is renamed as a prediction, and the self-citations ([17], [37], [55]) are used only as background examples, not as load-bearing external support. The one genuine circular step is the formal centerpiece, Eq. (6). There 𝝓*(𝜽) is defined as the argmin over 𝝓 of Lgen(𝝓; 𝜽, 𝒘*(𝜽)), while 𝒘*(𝜽) is defined as the argmin over 𝒘 of Ldro(𝒘; 𝜽, 𝝓). Since Ldro depends on 𝝓, the inner solution is not a function of 𝜽 alone; it depends on the middle-level variable that the middle-level problem is optimizing, making the constraint system self-referential rather than a well-posed multilevel program. The paper does not analyze this fixed-point structure, and the experiments do not exercise Eq. (6): the results are qualitative screen captures of LLM-only, static RAG, continual RAG, and a travel-planning prototype, with no numeric metrics or ablations. Thus the central claim that the multilevel formulation provides 'principled coordination' is partially circular as written, while the rest of the framework retains independent content, giving a score of 6 rather than a higher one.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The paper relies on standard optimization theory and domain-specific assumptions about urban data and RAG. It introduces no new physical entities. The main unstated load-bearing premise is that the multilevel optimization is practically solvable and beneficial, which is never demonstrated.

free parameters (3)
  • rho (KL divergence bound for uncertainty set U)
    Chosen by hand to control the allowable shift from the training distribution; value not specified.
  • epsilon (KL bound on domain weights w)
    Introduced to prevent over-concentration in the DRO formulation; value not specified.
  • K (number of retrieved documents)
    Top-K retrieval parameter used throughout; value not specified.
assumptions (6)
  • standard math Bilevel and multilevel optimization theory, including the existence and properties of nested optima.
    Invoked in Section 3.1 when presenting Eq. (1) and later in Eqs. (4)-(6).
  • standard math Distributionally robust optimization with KL divergence as the ambiguity set.
    Used in Section 3.5 to define the uncertainty set U and the DRO formulation in Eq. (6).
  • domain assumption Urban data is non-stationary and heterogeneous, requiring continual adaptation.
    Stated in Sections 1.1 and 2.1 as the motivation for the entire framework.
  • domain assumption RAG improves factual accuracy and adaptability of LLMs.
    Assumed throughout, supported by cited works [15, 80, 82] in Section 1.2.
  • ad hoc to paper The MoE architecture aligns with the multilevel optimization hierarchy.
    Claimed in Section 3.4 without a formal mapping or proof; the paper states the alignment is 'natural'.
  • ad hoc to paper The multi-timescale update scheduling can be inverted per domain without loss of performance.
    In Section 2.1 the paper acknowledges the hierarchy can invert for traffic or education, which makes the temporal claims flexible and difficult to falsify.

how reviews work

0 comments
Cite this review

Pith. "Pith review of UrbanMind: Towards Urban General Intelligence via Tool-Enhanced Retrieval-Augmented Generation and Multilevel Optimization." pith.science (2026). https://pith.science/paper/ZQAENLDU

@misc{pith2026250704706,
  author       = {Pith},
  title        = {Pith review of: UrbanMind: Towards Urban General Intelligence via Tool-Enhanced Retrieval-Augmented Generation and Multilevel Optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZQAENLDU}},
  note         = {Machine review of arXiv:2507.04706}
}
read the original abstract

Urban general intelligence (UGI) refers to the capacity of AI systems to autonomously perceive, reason, and act within dynamic and complex urban environments. In this paper, we introduce UrbanMind, a tool-enhanced retrieval-augmented generation (RAG) framework designed to facilitate UGI. Central to UrbanMind is a novel architecture based on Continual Retrieval-Augmented MoE-based LLM (C-RAG-LLM), which dynamically incorporates domain-specific knowledge and evolving urban data to support long-term adaptability. The architecture of C-RAG-LLM aligns naturally with a multilevel optimization framework, where different layers are treated as interdependent sub-problems. Each layer has distinct objectives and can be optimized either independently or jointly through a hierarchical learning process. The framework is highly flexible, supporting both end-to-end training and partial layer-wise optimization based on resource or deployment constraints. To remain adaptive under data drift, it is further integrated with an incremental corpus updating mechanism. Evaluations on real-world urban tasks of a variety of complexity verify the effectiveness of the proposed framework. This work presents a promising step toward the realization of general-purpose LLM agents in future urban environments.

Figures

Figures reproduced from arXiv: 2507.04706 by the authors.

Figure 1
Figure 1. Tool-Enhanced RAG with Continual Learning [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Naive RAG Pipeline Workflow retrieved chunks provide external context that is critical for grounding the model’s reasoning in factual knowledge. Finally, in the generation phase, the retrieved chunks are combined with the original query and fed into a large language model, which generates a response by leveraging both the query and the retrieved knowledge. 2.3 Urban Intelligence Tasks Urban intelligence tasks encomp… view at source ↗
Figure 3
Figure 3. UrbanMind Framework domains, allowing the retrieval module to perform targeted, context-aware searches. Retrieved entries R𝑡 (𝑥𝑡) are filtered based on relevance scores and uncertainty estimates before being passed to the integration module for downstream processing. Given the non-stationary nature of urban environments, the continual update of the knowledge base K𝑡 is critical for maintaining retrieval accuracy and… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: UrbanMind over Cloud-Edge Architecture 3 Multilevel Optimization Strategy for UrbanMind UrbanMind adopts a multilevel optimization strategy that aligns naturally with its modular design and integrates seamlessly with MoE architectures. This approach enables coordinated…
Figure 6
Figure 6. Figure 6: LLM,Static RAG LLM and Continual RAG Com [PITH_FULL_IMAGE:figures/full_fig_p018_6.png]
Figure 8
Figure 8. Figure 8: Tool-Enhanced UrbanMind Experimental Framework To travel from Anting Town to Hongqiao Airport, you have several options: 1. **Subway**: Take Line 11 from Anting Ecopark Station to Hongqiao Airport T1/T2 Station. This journey takes about 30-40 minutes and is the most co…
Figure 7
Figure 7. Figure 7: LLM,Static RAG LLM and Continual RAG Com [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 10
Figure 10. Figure 10: Travel Plan with RAG-LLM I want to travel to HongQiao Airport from Anting Town, how to go there? Given your current situation at 23:00, here are your options to travel to HongQiao Airport from Anting Town: 1. **Subway:** The last train of Line 11 has already departed …

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

94 extracted references · 60 canonical work pages

  1. [30]

    GeoFabrik GmbH. 2025. OpenStreetMap Data Extracts for China. https://download.geofabrik.de/asia/china.html. Accessed: 2025-04-29

    work page 2025

  2. [1]

    Ali Ben Abbes, ImedRiadh Farah, and Vincent Barra. 2016. Urban growth analysis using multi-temporal satellite images, non-stationary decomposition methods and stochastic modeling. World Acad Sci Eng Technol Int J Comput Electr Autom Control Inf Eng 10, 10 (2016), 1791–1797

    2016

  3. [2]

    Asma Ait Ouallane, Assia Bakali, Ayoub Bahnasse, Said Broumi, and Mohamed Talea. 2022. Fusion of engineering insights and emerging trends: Intelligent urban traffic management system. Information Fusion 88 (2022), 218–248

    2022

  4. [3]

    Sarwat Ali and M Arif Wani. 2024. Tri-level Optimization for Gradient-based Neural Architecture Search. In 2024 International Conference on Machine Learning and Applications (ICMLA) . IEEE, 1546–1552

    2024

  5. [4]

    Pasquale Balsebre, Weiming Huang, Gao Cong, and Yi Li. 2023. Cityfm: City foundation models to solve urban challenges. arXiv preprint arXiv:2310.00583 (2023)

    work page Pith review arXiv 2023

  6. [5]

    Firas Bayram, Bestoun S Ahmed, and Andreas Kassler. 2022. From concept drift to model degradation: An overview on performance-aware drift detectors. Knowledge-Based Systems 245 (2022), 108632

    2022

  7. [6]

    Luís MA Bettencourt. 2021. Introduction to urban science: evidence and theory of cities as complex systems. (2021)

    2021

  8. [7]

    Simon Elias Bibri. 2021. Data-driven smart sustainable cities of the future: Urban computing and intelligence for strategic, short-term, and joined-up planning. Computational Urban Science 1, 1 (2021), 8

    2021

Show all 94 references
  1. [8]

    Magdalena Biesialska, Katarzyna Biesialska, and Marta R Costa-Jussa. 2020. Continual lifelong learning in natural language processing: A survey. arXiv preprint arXiv:2012.09823 (2020)

    2020 arXiv

  2. [9]

    Chang Chen, Min Li, Zhihua Wu, Dianhai Yu, and Chao Yang. 2022. Ta-moe: Topology-aware large scale mixture-of- expert training. Advances in Neural Information Processing Systems 35 (2022), 22173–22186

    2022

  3. [10]

    Xu Chen, Junshan Wang, and Kunqing Xie. 2021. TrafficStream: A streaming traffic flow forecasting framework based on graph neural networks and continual learning. arXiv preprint arXiv:2106.06273 (2021)

    2021 arXiv

  4. [11]

    Xingdi Chen, Yu Xiong, and Kai Yang. 2024. Robust beamforming for downlink multi-cell systems: A bilevel optimization perspective. In Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 38. 7969–7977

    2024

  5. [12]

    Zhuo Chen, Ruoxi Chen, and Songtao Chen. 2021. Intelligent management information system of urban planning based on GIS. Journal of Intelligent & Fuzzy Systems 40, 4 (2021), 6007–6016

    2021

  6. [13]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human...

    2019

  7. [14]

    Beniamino di Martino, Domenico Di Sivo, and Alba Amato. 2025. Cloud, Edge, and Mobile Computing: Synergies for the Future of Smart Cities. In International Conference on Advanced Information Networking and Applications . Springer, 158–166

    2025

  8. [15]

    Wenhao Ding, Yulong Cao, Ding Zhao, Chaowei Xiao, and Marco Pavone. 2024. Realgen: Retrieval augmented generation for controllable traffic scenarios. In European Conference on Computer Vision . Springer, 93–110

    2024

  9. [16]

    Fei Dou, Jin Ye, Geng Yuan, Qin Lu, Wei Niu, Haijian Sun, Le Guan, Guoyu Lu, Gengchen Mai, Ninghao Liu, et al. 2023. Towards artificial general intelligence (agi) in the internet of things (iot): Opportunities and challenges. arXiv preprint arXiv:2309.07438 (2023). J. ACM, Vol...

    2023 arXiv

  10. [17]

    Shaoyu Dou, Kai Yang, Yang Jiao, Chengbo Qiu, and Kui Ren. 2024. Anomaly Detection in Event-triggered Traffic Time Series via Similarity Learning. IEEE Transactions on Dependable and Secure Computing (2024)

    2024

  11. [18]

    Nan Du, Yanping Huang, Andrew M Dai, Simon Tong, Dmitry Lepikhin, Yuanzhong Xu, Maxim Krikun, Yanqi Zhou, Adams Wei Yu, Orhan Firat, et al . 2022. Glam: Efficient scaling of language models with mixture-of-experts. In International conference on machine learning . PMLR, 5547–5569

    2022

  12. [19]

    Cristian Duran-Mateluna et al. 2025. Adaptive Robust Optimization Models for DER Planning in Distribution Networks under Long-and Short-Term Uncertainties. arXiv e-prints (2025), arXiv–2503

    2025

  13. [20]

    Fernando Elizalde-Ramírez, Romeo Sanchez Nigenda, Iris A Martínez-Salazar, and Yasmín Á Ríos-Solís. 2019. Travel plans in public transit networks using artificial intelligence planning models. Applied Artificial Intelligence 33, 5 (2019), 440–461

    2019

  14. [21]

    Mohammed A Fadhel, Ali M Duhaim, Ahmed Saihood, Ahmed Sewify, Mokhaled NA Al-Hamadani, AS Albahri, Laith Alzubaidi, Ashish Gupta, Sayedali Mirjalili, and Yuantong Gu. 2024. Comprehensive systematic review of information fusion methods in smart cities and urban environments. In...

    2024

  15. [22]

    Yuxin Fan, Yuxiang Wang, Lipeng Liu, Xirui Tang, Na Sun, and Zidong Yu. 2025. Research on the Online Update Method for Retrieval-Augmented Generation (RAG) Model with Incremental Learning. arXiv preprint arXiv:2501.07063 (2025)

    2025 arXiv

  16. [23]

    Nanyi Fei, Zhiwu Lu, Yizhao Gao, Guoxing Yang, Yuqi Huo, Jingyuan Wen, Haoyu Lu, Ruihua Song, Xin Gao, Tao Xiang, et al. 2022. Towards artificial general intelligence via a multimodal foundation model. Nature Communications 13, 1 (2022), 3094

    2022

  17. [24]

    Luciano Floridi and Massimo Chiriatti. 2020. GPT-3: Its nature, scope, limits, and consequences. Minds and Machines 30 (2020), 681–694

    2020

  18. [25]

    Francis Fortin, Julie Delle Donne, and Justine Knop. 2021. The use of social media in intelligence and its impact on police work. Policing in an Age of Reform: An Agenda for Research and Practice (2021), 213–231

    2021

  19. [26]

    Cátia AR Freire, Fernando AF Ferreira, Elias G Carayannis, and João JM Ferreira. 2021. Artificial intelligence and smart cities: A DEMATEL approach to adaptation challenges and initiatives. IEEE Transactions on Engineering Management 70, 5 (2021), 1881–1899

    2021

  20. [27]

    Trevor Gale, Deepak Narayanan, Cliff Young, and Matei Zaharia. 2023. Megablocks: Efficient sparse training with mixture-of-experts. Proceedings of Machine Learning and Systems 5 (2023), 288–304

    2023

  21. [28]

    Daniele Gammelli, James Harrison, Kaidi Yang, Marco Pavone, Filipe Rodrigues, and Francisco C Pereira. 2023. Graph reinforcement learning for network control via bi-level optimization. arXiv preprint arXiv:2305.09129 (2023)

    2023 arXiv

  22. [29]

    Qiang Gao, Zhipeng Luo, Diego Klabjan, and Fengli Zhang. 2022. Efficient architecture search for continual learning. IEEE Transactions on Neural Networks and Learning Systems 34, 11 (2022), 8555–8565

    2022

  23. [31]

    Ben Goertzel. 2014. Artificial general intelligence: concept, state of the art, and future prospects. Journal of Artificial General Intelligence 5, 1 (2014), 1

    2014

  24. [32]

    Fatemeh Golpayegani, Saeedeh Ghanadbashi, and Maha Riad. 2021. Urban emergency management using intelligent traffic systems: challenges and future directions. In 2021 IEEE International Smart Cities Conference (ISC2) . IEEE, 1–4

    2021

  25. [33]

    Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948 (2025)

    2025 arXiv

  26. [34]

    Shanshan Han, Qifan Zhang, Yuhang Yao, Weizhao Jin, Zhaozhuo Xu, and Chaoyang He. 2024. LLM multi-agent systems: Challenges and open problems. arXiv preprint arXiv:2402.03578 (2024)

    2024 arXiv

  27. [35]

    Zijian Hei, Weiling Liu, Wenjie Ou, Juyi Qiao, Junming Jiao, Guowen Song, Ting Tian, and Yi Lin. 2024. Dr-rag: Applying dynamic document relevance to retrieval-augmented generation for question-answering. arXiv preprint arXiv:2406.07348 (2024)

    2024 arXiv

  28. [36]

    Zhongzhen Huang, Kui Xue, Yongqi Fan, Linjie Mu, Ruoyu Liu, Tong Ruan, Shaoting Zhang, and Xiaofan Zhang. 2024. Tool Calling: Enhancing Medication Consultation via Retrieval-Augmented Large Language Models. arXiv preprint arXiv:2404.17897 (2024)

    2024 arXiv

  29. [37]

    Chengtao Jian, Kai Yang, and Yang Jiao. 2024. Tri-Level Navigator: LLM-Empowered Tri-Level Learning for Time Series OOD Generalization. Advances in Neural Information Processing Systems 37 (2024), 110613–110642

    2024

  30. [38]

    Yang Jiao, Kai Yang, Dongjing Song, and Dacheng Tao. 2022. Timeautoad: Autonomous anomaly detection with self-supervised contrastive loss for multivariate time series. IEEE Transactions on Network Science and Engineering 9, 3 (2022), 1604–1619

    2022

  31. [39]

    Yang Jiao, Kai Yang, Tiancheng Wu, Dongjin Song, and Chengtao Jian. [n. d.]. Asynchronous Distributed Bilevel Optimization. In The Eleventh International Conference on Learning Representations . J. ACM, Vol. 37, No. 4, Article 111. Publication date: August 2018. 111:22 Kai Yan...

    2018

  32. [40]

    Yang Jiao, Kai Yang, Tiancheng Wu, Dongjin Song, and Chengtao Jian. 2022. Asynchronous Distributed Bilevel Optimization. In The Eleventh International Conference on Learning Representations

    2022

  33. [41]

    Shaheen Khatoon, Amna Asif, Md Maruf Hasan, and Majed Alshamari. 2022. Social media-based intelligence for disaster response and management in smart cities. In Artificial Intelligence, Machine Learning, and Optimization Tools for Smart Cities: Designing for Sustainability . Sp...

    2022

  34. [42]

    Minyoung Kim and Timothy Hospedales. 2025. A Stochastic Approach to Bi-Level Optimization for Hyperparameter Optimization and Meta Learning. In Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 39. 17913–17920

    2025

  35. [43]

    Karima Kourtit. 2021. City intelligence for enhancing urban performance value: a conceptual study on data decompo- sition in smart cities. Asia-Pacific Journal of Regional Science 5, 1 (2021), 191–222

    2021

  36. [44]

    Ehsan Latif, Gengchen Mai, Matthew Nyaaba, Xuansheng Wu, Ninghao Liu, Guoyu Lu, Sheng Li, Tianming Liu, and Xiaoming Zhai. 2023. Artificial general intelligence (AGI) for education. arXiv preprint arXiv:2304.12479 1 (2023)

    2023 arXiv

  37. [45]

    Namkyeong Lee, Edward De Brouwer, Ehsan Hajiramezanali, Tommaso Biancalani, Chanyoung Park, and Gabriele Scalia. 2025. RAG-Enhanced Collaborative LLM Agents for Drug Discovery. arXiv preprint arXiv:2502.17506 (2025)

    2025

  38. [46]

    Timothée Lesort, Vincenzo Lomonaco, Andrei Stoian, Davide Maltoni, David Filliat, and Natalia Díaz-Rodríguez. 2020. Continual learning for robotics: Definition, framework, learning strategies, opportunities and challenges. Information fusion 58 (2020), 52–68

    2020

  39. [47]

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, et al. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in neural information processing s...

    2020

  40. [48]

    Xinzhe Li. 2025. A Review of Prominent Paradigms for LLM-Based Agents: Tool Use, Planning (Including RAG), and Feedback Learning. In Proceedings of the 31st International Conference on Computational Linguistics . 9760–9779

    2025

  41. [49]

    Xiaopeng Li, Pengyue Jia, Derong Xu, Yi Wen, Yingyi Zhang, Wenlin Zhang, Wanyu Wang, Yichao Wang, Zhaocheng Du, Xiangyang Li, et al. 2025. A Survey of Personalization: From RAG to Agent. arXiv preprint arXiv:2504.10147 (2025)

    2025 arXiv

  42. [50]

    Xiang Li, Lin Zhao, Lu Zhang, Zihao Wu, Zhengliang Liu, Hanqi Jiang, Chao Cao, Shaochen Xu, Yiwei Li, Haixing Dai, et al. 2024. Artificial general intelligence for medical imaging analysis. IEEE Reviews in Biomedical Engineering (2024)

    2024

  43. [51]

    Huimin Liu, Qingming Zhan, Sihang Gao, and Chen Yang. 2019. Seasonal variation of the spatially non-stationary association between land surface temperature and urban landscape. Remote Sensing 11, 9 (2019), 1016

    2019

  44. [52]

    Risheng Liu, Jiaxin Gao, Jin Zhang, Deyu Meng, and Zhouchen Lin. 2021. Investigating bi-level optimization for learning and vision from a unified perspective: A survey and beyond. IEEE Transactions on Pattern Analysis and Machine Intelligence 44, 12 (2021), 10045–10067

    2021

  45. [53]

    Yu Liu, Jingtao Ding, and Yong Li. 2023. Knowsite: Leveraging urban knowledge graph for site selection. InProceedings of the 31st ACM International Conference on Advances in Geographic Information Systems . 1–12

    2023

  46. [54]

    Paul A Longley and Carolina Tobón. 2004. Spatial dependence and heterogeneity in patterns of hardship: an intra-urban analysis. Annals of the Association of American Geographers 94, 3 (2004), 503–519

    2004

  47. [55]

    Hui Ma and Kai Yang. 2023. Metastnet: Multimodal meta-learning for cellular traffic conformal prediction. IEEE Transactions on Network Science and Engineering 11, 2 (2023), 1999–2011

    2023

  48. [56]

    Hui Ma, Kai Yang, and Man-On Pun. 2023. Cellular traffic prediction via deep state space models with attention mechanism. Computer Communications 197 (2023), 276–283

    2023

  49. [57]

    Vinod Mahor, Romil Rawat, Anil Kumar, Bhagwati Garg, Kiran Pachlasiya, et al. 2023. IoT and artificial intelligence techniques for public safety and security. In Smart urban computing applications . River Publishers, 111–126

    2023

  50. [58]

    María Luisa Menéndez, Julio Angel Pardo, Leandro Pardo, and María del C Pardo. 1997. The jensen-shannon divergence. Journal of the Franklin Institute 334, 2 (1997), 307–318

    1997

  51. [59]

    Meredith Ringel Morris, Jascha Sohl-Dickstein, Noah Fiedel, Tris Warkentin, Allan Dafoe, Aleksandra Faust, Clement Farabet, and Shane Legg. 2024. Position: Levels of AGI for operationalizing progress on the path to AGI. In Forty-first International Conference on Machine Learning

    2024

  52. [60]

    Mahima Nama, Ankita Nath, Nancy Bechra, Jitendra Bhatia, Sudeep Tanwar, Manish Chaturvedi, and Balqies Sadoun

  53. [61]

    Cuong V Nguyen, Alessandro Achille, Michael Lam, Tal Hassner, Vijay Mahadevan, and Stefano Soatto. 2019. Toward understanding catastrophic forgetting in continual learning. arXiv preprint arXiv:1908.01091 (2019)

    2019 arXiv

  54. [62]

    Tarannum Nisha, Duong Tung Nguyen, and Vijay K Bhargava. 2022. A bilevel programming framework for joint edge resource management and pricing. IEEE Internet of Things Journal 9, 18 (2022), 17280–17291

    2022

  55. [63]

    Nelson Pacheco Rocha, Ana Dias, Gonçalo Santinha, Mário Rodrigues, Carlos Rodrigues, Alexandra Queirós, Rute Bastardo, and João Pavão. 2022. Systematic literature review of context-awareness applications supported by smart cities’ infrastructures. SN Applied Sciences 4, 4 (2022), 90

    2022

  56. [64]

    Shishir G Patil, Tianjun Zhang, Xin Wang, and Joseph E Gonzalez. 2024. Gorilla: Large language model connected with massive apis. Advances in Neural Information Processing Systems 37 (2024), 126544–126565. J. ACM, Vol. 37, No. 4, Article 111. Publication date: August 2018. Urb...

    2024

  57. [65]

    Jary Pomponi, Simone Scardapane, Vincenzo Lomonaco, and Aurelio Uncini. 2020. Efficient continual learning in neural networks with embedding regularization. Neurocomputing 397 (2020), 139–148

    2020

  58. [66]

    Nihal Poredi, Yu Chen, Xiaohua Li, and Erik Blasch. 2023. Enhance public safety surveillance in smart cities by fusing optical and thermal cameras. In 2023 26th International Conference on Information Fusion (FUSION) . IEEE, 1–7

    2023

  59. [67]

    Qi Qian, Shenghuo Zhu, Jiasheng Tang, Rong Jin, Baigui Sun, and Hao Li. 2019. Robust optimization over multiple domains. In Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 33. 4739–4746

    2019

  60. [68]

    Haoxuan Qu, Hossein Rahmani, Li Xu, Bryan Williams, and Jun Liu. 2025. Recent advances of continual learning in computer vision: An overview. IET Computer Vision 19, 1 (2025), e70013

    2025

  61. [69]

    Kadiyala Ramana, Gautam Srivastava, Madapuri Rudra Kumar, Thippa Reddy Gadekallu, Jerry Chun-Wei Lin, Mamoun Alazab, and Celestine Iwendi. 2023. A vision transformer approach for traffic congestion prediction in urban areas. IEEE Transactions on Intelligent Transportation Syst...

    2023

  62. [70]

    Roopa Ravish and Shanta Ranga Swamy. 2021. Intelligent traffic management: A review of challenges, solutions, and future perspectives. Transport and Telecommunication 22, 2 (2021), 163–182

    2021

  63. [71]

    Sujoy Roychowdhury, Sumit Soman, HG Ranjani, Neeraj Gunda, Vansh Chhabra, and Sai Krishna Bala. 2024. Evaluation of rag metrics for question answering in the telecom domain. arXiv preprint arXiv:2407.12873 (2024)

    2024 arXiv

  64. [72]

    Xue Rui, Ziqiang Li, Yang Cao, Ziyang Li, and Weiguo Song. 2023. DILRS: Domain-incremental learning for semantic segmentation in multi-source remote sensing data. Remote Sensing 15, 10 (2023), 2541

    2023

  65. [73]

    Ludger Rüschendorf. 1985. The Wasserstein distance and approximation theorems. Probability Theory and Related Fields 70, 1 (1985), 117–129

    1985

  66. [74]

    Shweta Srivastava, Aditya Bisht, and Neetu Narayan. 2017. Safety and security in smart cities using artificial intelli- gence—A review. In 2017 7th international conference on cloud computing, data science & engineering-confluence . IEEE, 130–133

    2017

  67. [75]

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

    2023 arXiv

  68. [76]

    Liyuan Wang, Xingxing Zhang, Hang Su, and Jun Zhu. 2024. A comprehensive survey of continual learning: Theory, method and application. IEEE Transactions on Pattern Analysis and Machine Intelligence (2024)

    2024

  69. [77]

    Liyuan Wang, Xingxing Zhang, Kuo Yang, Longhui Yu, Chongxuan Li, Lanqing Hong, Shifeng Zhang, Zhenguo Li, Yi Zhong, and Jun Zhu. 2022. Memory replay with data compression for continual learning.arXiv preprint arXiv:2202.06592 (2022)

    2022 arXiv

  70. [78]

    Shuhe Wang, Shengyu Zhang, Jie Zhang, Runyi Hu, Xiaoya Li, Tianwei Zhang, Jiwei Li, Fei Wu, Guoyin Wang, and Eduard Hovy. 2024. Reinforcement learning enhanced llms: A survey. arXiv preprint arXiv:2412.10400 (2024)

    2024 arXiv

  71. [79]

    Shirley Wu, Shiyu Zhao, Qian Huang, Kexin Huang, Michihiro Yasunaga, Kaidi Cao, Vassilis Ioannidis, Karthik Subbian, Jure Leskovec, and James Y Zou. 2024. Avatar: Optimizing llm agents for tool usage via contrastive reasoning.Advances in Neural Information Processing Systems 3...

    2024

  72. [80]

    Peng Xia, Kangyu Zhu, Haoran Li, Hongtu Zhu, Yun Li, Gang Li, Linjun Zhang, and Huaxiu Yao. 2024. Rule: Reliable multimodal rag for factuality in medical vision language models. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing . 1081–1093

    2024

  73. [81]

    Fengli Xu, Jun Zhang, Chen Gao, Jie Feng, and Yong Li. 2023. Urban Generative Intelligence (UGI): A Foundational Platform for Agents in Embodied City Environment. CoRR (2023)

    2023

  74. [82]

    Zhentao Xu, Mark Jerome Cruz, Matthew Guevara, Tie Wang, Manasi Deshpande, Xiaofeng Wang, and Zheng Li. 2024. Retrieval-augmented generation with knowledge graphs for customer service question answering. In Proceedings of the 47th International ACM SIGIR Conference on Research...

    2024

  75. [83]

    Fei Yang, Yixin Hua, Xiang Li, Zhenkai Yang, Xinkai Yu, and Teng Fei. 2022. A survey on multisource heterogeneous urban sensor access and data management technologies. Measurement: Sensors 19 (2022), 100061

    2022

  76. [84]

    Kai Yang, Jianwei Huang, Yihong Wu, Xiaodong Wang, and Mung Chiang. 2014. Distributed robust optimization (DRO), part I: Framework and example. Optimization and Engineering 15 (2014), 35–67

    2014

  77. [85]

    Kai Yang, Yihong Wu, Jianwei Huang, Xiaodong Wang, and Sergio Verdú. 2008. Distributed robust optimization for communication networks. In IEEE INFOCOM 2008-The 27th Conference on Computer Communications. IEEE, 1157–1165

    2008

  78. [86]

    Li Yang, Zhipeng Luo, Shiming Zhang, Fei Teng, and Tianrui Li. 2024. Continual Learning for Smart City: A Survey. IEEE Transactions on Knowledge and Data Engineering (2024)

    2024

  79. [87]

    Min Yang, Wenting Tu, Qiang Qu, Kai Lei, Xiaojun Chen, Jia Zhu, and Ying Shen. 2019. MARES: multitask learning algorithm for Web-scale real-time event summarization. World Wide Web 22 (2019), 499–515

    2019

  80. [88]

    Masoumeh Zareapoor, Pourya Shamsolmoali, and Fateme Vesaghati. 2024. Efficient Routing in Sparse Mixture-of- Experts. In 2024 International Joint Conference on Neural Networks (IJCNN) . IEEE, 1–8

    2024

  81. [89]

    Han Zhang, Yu Lei, Lin Gui, Min Yang, Yulan He, Hui Wang, and Ruifeng Xu. 2024. Cppo: Continual learning for reinforcement learning with human feedback. In The Twelfth International Conference on Learning Representations . J. ACM, Vol. 37, No. 4, Article 111. Publication date:...

    2024

  82. [90]

    Rongyu Zhang, Yun Chen, Chenrui Wu, Fangxin Wang, and Bo Li. 2024. Multi-level personalized federated learning on heterogeneous and long-tailed data. IEEE Transactions on Mobile Computing 23, 12 (2024), 12396–12409

    2024

  83. [91]

    Weijia Zhang, Jindong Han, Zhao Xu, Hang Ni, Hao Liu, and Hui Xiong. 2024. Urban foundation models: A survey. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 6633–6643

    2024

  84. [92]

    Siyun Zhao, Yuqing Yang, Zilong Wang, Zhiyuan He, Luna K Qiu, and Lili Qiu. 2024. Retrieval augmented generation (rag) and beyond: A comprehensive survey on how to make your llms use external data more wisely. arXiv preprint arXiv:2409.14924 (2024)

    2024 arXiv

  85. [93]

    Yuyue Zhao, Jiancan Wu, Xiang Wang, Wei Tang, Dingxian Wang, and Maarten De Rijke. 2024. Let me do it for you: Towards llm empowered recommendation via tool learning. In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retri...

    2024

  86. [2021]

    International Journal of Communication Systems 34, 9 (2021), e4814

    Machine learning-based traffic scheduling techniques for intelligent transportation system: Opportunities and challenges. International Journal of Communication Systems 34, 9 (2021), e4814

    2021

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.