pith. sign in

arxiv: 2606.17637 · v1 · pith:34YZRCELnew · submitted 2026-06-16 · 💻 cs.AI

Brick-DICL: Dynamic In-Context Learning for Automated Brick Schema Classification

Yiyue Qian , Shinan Zhang , Huan Song , Negin Sokhandan , Hannah Marlowe , Diego Socolinsky This is my paper

Pith reviewed 2026-06-27 01:15 UTC · model grok-4.3

classification 💻 cs.AI
keywords Brick schemadynamic in-context learningretrieval augmented generationbuilding management systemsautomated classificationlarge language modelsontology mappingBMS points
0
0 comments X

The pith

Brick-DICL uses two-stage RAG retrieval inside dynamic in-context learning to map building points to any of 936 Brick classes.

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

The paper presents Brick-DICL as a way to classify building management system metadata into the Brick ontology without site-specific fine-tuning. It retrieves similar metadata examples to supply domain context and narrows the set of candidate classes before asking an LLM for a label, then cross-checks outputs from several models to flag cases for human review. If the method works, it removes the main barriers of class volume, LLM knowledge gaps, and manual labor that currently slow standardization of building data. A reader would care because standardized mappings enable easier integration of energy and operations data across equipment from different makers.

Core claim

Brick-DICL is a two-stage dynamic in-context learning framework that first applies metadata-RAG to retrieve relevant examples and then applies class-RAG to restrict the candidate set among 936 Brick classes, combined with multi-LLM filtering that routes low-confidence predictions to human review, and this combination produces higher classification accuracy than prior methods on multiple building datasets.

What carries the argument

The two-stage RAG pipeline (metadata-RAG for example retrieval and class-RAG for candidate narrowing) inside a dynamic in-context learning loop, plus the multi-LLM consensus filter.

If this is right

  • The framework applies to building management systems from any manufacturer and any metadata format.
  • Multi-LLM filtering reduces the volume of cases that need manual verification.
  • The method speeds up the process of bringing building data into a standardized, interoperable form.
  • Accuracy gains hold across diverse building datasets without requiring domain-specific training data.

Where Pith is reading between the lines

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

  • The same retrieval-plus-filter pattern could be tested on classification tasks that use other large, hierarchical building or industrial ontologies.
  • Gains in base LLM knowledge or retrieval quality could shrink the fraction of cases sent for human review.
  • Widespread adoption would support larger-scale energy optimization applications that rely on consistent point labels across sites.

Load-bearing premise

A small number of retrieved metadata examples is enough to give an LLM the domain knowledge required to choose correctly among 936 Brick classes.

What would settle it

Evaluation on a new building dataset where Brick-DICL accuracy falls below the best baseline method or where most outputs still require human review.

Figures

Figures reproduced from arXiv: 2606.17637 by Diego Socolinsky, Hannah Marlowe, Huan Song, Negin Sokhandan, Shinan Zhang, Yiyue Qian.

Figure 1
Figure 1. Figure 1: An example of Brick schema hierarchy, illustrating relationships among equipment, locations, and point classes in [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The framework of Brick-DICL: (a) Raw data is first preprocessed to extract and standardize meta-data and candidate [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Stage2 Hits@1 and Hits@3 accuracy comparison for BMS point classification on B1 and B2 buildings. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Impact of hyperparameters on Brick-DICL. Left two: number of shots in metadata ICL. Right two: number of retrieved [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of four filtering strategies among [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Performance comparison among model variants on [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
read the original abstract

Building Management Systems (BMS) are essential for optimizing energy efficiency and operational performance in modern buildings. However, the lack of standardization across BMS points from different manufacturers creates significant barriers to integration and data utilization. While the Brick schema offers a standardized ontology for building systems, mapping BMS points to appropriate Brick classes presents three critical challenges: (i) the extensive number of Brick classes (936 in the latest version), (ii) limited domain-specific knowledge in large language models (LLMs), and (iii) substantial manual effort required for verification. To address these challenges, we propose Brick-DICL, a two-stage dynamic in-context learning framework for automated Brick schema classification. Brick-DICL consists of two primary components: metadata-RAG, which retrieves relevant examples to enhance LLMs' domain knowledge, and class-RAG, which narrows down potential Brick classes to address the large classification space. Additionally, we implement a multi-LLM filtering mechanism that compares predictions across multiple models, flagging low-confidence classifications for human review. As a result: (i) General: Brick-DICL is applicable to any building management system regardless of manufacturer or metadata format; (ii) Novel and Powerful: as the first dynamic in-context learning approach for Brick schema classification, Brick-DICL achieves significant classification accuracy improvements on building datasets, outperforming existing methods; (iii) Efficient: our multi-LLM filtering strategy reduces manual verification effort, enabling rapid digital building onboarding. Extensive experiments demonstrate Brick-DICL's effectiveness across diverse building datasets, accelerating the path toward standardized, interoperable building management systems.

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.

Referee Report

2 major / 0 minor

Summary. The paper proposes Brick-DICL, a two-stage dynamic in-context learning framework for automated mapping of building management system (BMS) points to Brick schema classes. It combines metadata-RAG to supply domain examples to LLMs, class-RAG to narrow the 936-class space, and a multi-LLM filtering step that flags low-confidence predictions for human review. The authors claim the method is manufacturer-agnostic, achieves significant accuracy gains over prior methods on building datasets, and reduces manual verification effort.

Significance. If the empirical results hold, the work addresses a practical interoperability barrier in BMS by leveraging RAG-based in-context learning to handle a large ontology without fine-tuning. This could reduce onboarding time for standardized building data. The engineering framing (general applicability, multi-LLM consensus) is sensible, though the novelty claim as the first dynamic in-context approach for this task would benefit from explicit positioning against prior RAG or LLM-based schema mapping efforts.

major comments (2)
  1. [Abstract] Abstract: the central claim that Brick-DICL 'achieves significant classification accuracy improvements on building datasets, outperforming existing methods' is asserted without any reported accuracy numbers, dataset sizes or characteristics, baseline methods, or comparison tables. This absence prevents evaluation of the claim or of generalization across the 936 classes.
  2. [Abstract] Abstract (method description): the sufficiency of metadata-RAG plus class-RAG for overcoming LLM domain-knowledge gaps is presented as the core mechanism, yet no retrieval metrics (e.g., recall@K for ground-truth class inclusion, embedding similarity distributions) or ablation results isolating the RAG components are supplied. Without these, it is impossible to verify whether the narrowed candidate set reliably contains the correct class for the majority of inputs.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed review and constructive comments on the abstract. Both points identify legitimate gaps in how the abstract presents our claims and supporting evidence. We will revise the abstract in the next version to incorporate key quantitative results and references to supporting analyses from the experiments section.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that Brick-DICL 'achieves significant classification accuracy improvements on building datasets, outperforming existing methods' is asserted without any reported accuracy numbers, dataset sizes or characteristics, baseline methods, or comparison tables. This absence prevents evaluation of the claim or of generalization across the 936 classes.

    Authors: We agree that the abstract should provide concrete numbers to substantiate the claim. The full manuscript contains the experimental results (accuracy figures, dataset descriptions, baselines, and tables), but the abstract summarizes them at a high level for brevity. In the revision we will add specific accuracy values, dataset sizes/characteristics, and a brief reference to the baselines and comparison tables from Section 4. revision: yes

  2. Referee: [Abstract] Abstract (method description): the sufficiency of metadata-RAG plus class-RAG for overcoming LLM domain-knowledge gaps is presented as the core mechanism, yet no retrieval metrics (e.g., recall@K for ground-truth class inclusion, embedding similarity distributions) or ablation results isolating the RAG components are supplied. Without these, it is impossible to verify whether the narrowed candidate set reliably contains the correct class for the majority of inputs.

    Authors: The abstract describes the high-level mechanism without the supporting retrieval statistics or ablations. The experiments section of the manuscript reports overall performance and some component analysis, but does not include the specific retrieval metrics (recall@K, similarity distributions) or dedicated RAG ablations requested. We will revise the abstract to reference the available experimental evidence and, if space permits, add a short statement on retrieval effectiveness; a more complete set of retrieval metrics and ablations can be added to the results section if the referee deems it necessary. revision: partial

Circularity Check

0 steps flagged

No circularity: engineering framework with empirical results, no derivation chain present.

full rationale

The paper describes an applied two-stage RAG-based classification pipeline (metadata-RAG + class-RAG + multi-LLM filter) evaluated on building datasets. No equations, first-principles derivations, fitted parameters renamed as predictions, or self-citation chains appear in the method or claims. The accuracy improvements are presented as experimental outcomes on external benchmarks rather than reductions to the method's own inputs. This matches the default expectation of no significant circularity for non-derivational work.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Review is limited to the abstract; no explicit free parameters, axioms, or invented entities are detailed beyond the stated challenges of LLM domain knowledge and large class space.

axioms (2)
  • domain assumption Large language models possess limited domain-specific knowledge for Brick schema classification
    Listed as challenge (ii) in the abstract
  • domain assumption Retrieving relevant metadata examples can meaningfully augment LLM performance on this task
    Core premise of the metadata-RAG component

pith-pipeline@v0.9.1-grok · 5837 in / 1342 out tokens · 35033 ms · 2026-06-27T01:15:57.292671+00:00 · methodology

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

62 extracted references · 14 canonical work pages · 4 internal anchors

  1. [1]

    Many-shot in-context learning

    Rishabh Agarwal, Avi Singh, Lei Zhang, Bernd Bohnet, Luis Rosias, Stephanie Chan, Biao Zhang, Ankesh Anand, Zaheer Abbas, Azade Nova, et al. Many-shot in-context learning. In NeurIPS, volume 37, pages 76930–76966, 2024

    2024

  2. [2]

    Brick: Towards a unified metadata schema for buildings

    Bharathan Balaji, Arka Bhattacharya, Gabriel Fierro, Jingkun Gao, Joshua Gluck, Dezhi Hong, Aslak Johansen, Jason Koh, Joern Ploennigs, Yuvraj Agarwal, et al. Brick: Towards a unified metadata schema for buildings. In Proceedings of the 3rd ACM International Conference on Systems for Energy-Efficient Built Environments , pages 41–50, 2016

    2016

  3. [3]

    Brick: Metadata schema for portable smart building applications

    Bharathan Balaji, Arka Bhattacharya, Gabriel Fierro, Jingkun Gao, Joshua Gluck, Dezhi Hong, Aslak Johansen, Jason Koh, Joern Ploennigs, Yuvraj Agarwal, et al. Brick: Metadata schema for portable smart building applications. Applied energy, 226:1273–1292, 2018

    2018

  4. [4]

    Scibert: A pretrained language model for scientific text

    Iz Beltagy, Kyle Lo, and Arman Cohan. Scibert: A pretrained language model for scientific text. In EMNLP, pages 3615–3620, 2019

    2019

  5. [5]

    Short paper: Analyzing metadata schemas for buildings: The good, the bad, and the ugly

    Arka Bhattacharya, Joern Ploennigs, and David Culler. Short paper: Analyzing metadata schemas for buildings: The good, the bad, and the ugly. In Proceedings of the 2nd ACM International Conference on Embedded Systems for Energy-Efficient Built Environments, pages 33–34, 2015

    2015

  6. [6]

    Language models are few-shot learners

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. In NeurIPS, 2020

    2020

  7. [7]

    Meta-in-context learning in large language models

    Julian Coda-Forno, Marcel Binz, Zeynep Akata, Matt Botvinick, Jane Wang, and Eric Schulz. Meta-in-context learning in large language models. In NeurIPS, 2023

    2023

  8. [8]

    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 NAACL-HLT, pages 4171–4186, 2019

    2019

  9. [9]

    Metadata standardization in iot-based building energy management systems

    Ying Ding, Yi Liu, and Yong Wu. Metadata standardization in iot-based building energy management systems. Journal of Cleaner Production, 210:1234–1243, 2019

    2019

  10. [10]

    Auto- mated classification of building energy metadata using ontology-based natural language processing

    Kevin Fiorelli, Kaveh Dehghanpour, and Alberto Sangiovanni-Vincentelli. Auto- mated classification of building energy metadata using ontology-based natural language processing. Applied Energy, 314:118915, 2023

    2023

  11. [11]

    A data-driven meta-data inference framework for building automation systems

    Jingkun Gao, Joern Ploennigs, and Mario Berges. A data-driven meta-data inference framework for building automation systems. In Proceedings of the 2nd ACM International Conference on Embedded Systems for Energy-Efficient Built Environments, pages 23–32, 2015

    2015

  12. [12]

    Making pre-trained language models better few-shot learners

    Tianyu Gao, Adam Fisch, and Danqi Chen. Making pre-trained language models better few-shot learners. arXiv preprint arXiv:2012.15723, 2020

    work page arXiv 2012

  13. [13]

    Openagi: When llm meets domain experts

    Yingqiang Ge, Wenyue Hua, Kai Mei, Juntao Tan, Shuyuan Xu, Zelong Li, Yongfeng Zhang, et al. Openagi: When llm meets domain experts. In NeurIPS, volume 36, pages 5539–5568, 2023

    2023

  14. [14]

    Evaluation and mitigation of the limitations of large language models in clinical decision-making

    Paul Hager, Friederike Jungmann, Robbie Holland, Kunal Bhagat, Inga Hubrecht, Manuel Knauer, Jakob Vielhauer, Marcus Makowski, Rickmer Braren, Georgios Kaissis, et al. Evaluation and mitigation of the limitations of large language models in clinical decision-making. Nature medicine, 30(9):2613–2622, 2024

    2024

  15. [15]

    Towards auto- matic spatial verification of sensor placement in buildings

    Dezhi Hong, Jorge Ortiz, Kamin Whitehouse, and David Culler. Towards auto- matic spatial verification of sensor placement in buildings. In Proceedings of the 5th ACM Workshop on Embedded Systems For Energy-Efficient Buildings , pages 1–8, 2013

    2013

  16. [16]

    Integrating building automation systems for energy efficiency: A review

    Ibrahim Katib, Waleed Abdulla, and Mohammed Nasr. Integrating building automation systems for energy efficiency: A review. Renewable and Sustainable Energy Reviews, 59:1571–1581, 2016

    2016

  17. [17]

    Data integration challenges in smart building systems: A review

    Seungwon Lee, Hui Li, and Yong Weng. Data integration challenges in smart building systems: A review. Energy and Buildings, 140:241–252, 2017

    2017

  18. [18]

    Verification- guided context optimization for tool calling via hierarchical llms-as-editors.arXiv preprint arXiv:2512.13860, 2025

    Henger Li, Shuangjie You, Flavio Di Palo, Yiyue Qian, and Ayush Jain. Verification- guided context optimization for tool calling via hierarchical llms-as-editors.arXiv preprint arXiv:2512.13860, 2025

    work page arXiv 2025

  19. [19]

    RoBERTa: A Robustly Optimized BERT Pretraining Approach

    Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1907

  20. [20]

    Specgen: Automated generation of formal program specifications via large language models

    Lezhi Ma, Shangqing Liu, Yi Li, Xiaofei Xie, and Lei Bu. Specgen: Automated generation of formal program specifications via large language models. arXiv preprint arXiv:2401.08807, 2024

    work page arXiv 2024

  21. [21]

    Opbench: A graph benchmark to combat the opioid crisis

    Tianyi Ma, Yiyang Li, Yiyue Qian, Zheyuan Zhang, Zehong Wang, Chuxu Zhang, and Yanfang Ye. Opbench: A graph benchmark to combat the opioid crisis. arXiv preprint arXiv:2602.14602, 2026

    work page arXiv 2026

  22. [22]

    Non-monotonic autoregressive sequence model

    Tianyi Ma, Yiyue Qian, Yiyang Li, Zehong Wang, Yifan Ding, Zheyuan Zhang, Yan Liang, Chuxu Zhang, and Yanfang Ye. Non-monotonic autoregressive sequence model. In ICML, 2026

    2026

  23. [23]

    Llm-empowered class imbalanced graph prompt learning for online drug trafficking detection

    Tianyi Ma, Yiyue Qian, Zehong Wang, Zheyuan Zhang, Chuxu Zhang, and Yanfang Ye. Llm-empowered class imbalanced graph prompt learning for online drug trafficking detection. In Findings of ACL, 2025

    2025

  24. [24]

    Bhygnn+: Unsupervised representation learning for heterophilic hypergraphs

    Tianyi Ma, Yiyue Qian, Zehong Wang, Zheyuan Zhang, Chuxu Zhang, and Yanfang Ye. Bhygnn+: Unsupervised representation learning for heterophilic hypergraphs. arXiv preprint arXiv:2602.14919, 2026

    work page arXiv 2026

  25. [25]

    Hypergraph representation learning with adaptive broad- casting and receiving

    Tianyi Ma, Yiyue Qian, Zehong Wang, Zheyuan Zhang, Shinan Zhang, Chuxu Zhang, and Fanny Ye. Hypergraph representation learning with adaptive broad- casting and receiving. In ICDM, 2025

    2025

  26. [26]

    Hypergraph contrastive learning for drug trafficking community detection

    Tianyi Ma, Yiyue Qian, Chuxu Zhang, and Yanfang Ye. Hypergraph contrastive learning for drug trafficking community detection. In ICDM, 2023

    2023

  27. [27]

    Adaptive expansion for hypergraph learning

    Tianyi Ma, Yiyue Qian, Shinan Zhang, Chuxu Zhang, and Yanfang Ye. Adaptive expansion for hypergraph learning. arXiv preprint arXiv:2502.15564, 2025

    work page arXiv 2025

  28. [28]

    Autodata: A multi-agent system for open web data collection

    Tianyi Ma, Yiyue Qian, Zheyuan Zhang, Zehong Wang, Xiaoye Qian, Feifan Bai, Yifan Ding, Xuwei Luo, Shinan Zhang, Keerthiram Murugesan, et al. Autodata: A multi-agent system for open web data collection. In NeurIPS, 2025

    2025

  29. [29]

    Instruction Tuning with GPT-4

    Baolin Peng, Chunyuan Li, Pengcheng He, Michel Galley, and Jianfeng Gao. Instruction tuning with gpt-4. arXiv preprint arXiv:2304.03277, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  30. [30]

    Graph Representation Learning Techniques for the Combat Against Online Abusive Activity

    Yiyue Qian. Graph Representation Learning Techniques for the Combat Against Online Abusive Activity. University of Notre Dame, 2024

    2024

  31. [31]

    Universal ring-of-abusers detection via multi-modal heterogeneous graph learning

    Yiyue Qian, Philip Chen, Song Cui, and De Chen. Universal ring-of-abusers detection via multi-modal heterogeneous graph learning. 2023

    2023

  32. [32]

    Dual-level hypergraph contrastive learning with adaptive temperature enhancement

    Yiyue Qian, Tianyi Ma, Chuxu Zhang, and Yanfang Ye. Dual-level hypergraph contrastive learning with adaptive temperature enhancement. In Companion of Brick-DICL: Dynamic In-Context Learning for Automated Brick Schema Classification KDD RelKD ’26, Aug 10, 2026, Jeju, Korea WWW, pages 859–862, 2024

    2026

  33. [33]

    Adaptive graph enhance- ment for imbalanced multi-relation graph learning

    Yiyue Qian, Tianyi Ma, Chuxu Zhang, and Yanfang Ye. Adaptive graph enhance- ment for imbalanced multi-relation graph learning. In WSDM, 2025

    2025

  34. [34]

    Co-modality graph contrastive learning for imbalanced node classification

    Yiyue Qian, Chunhui Zhang, Yiming Zhang, Qianlong Wen, Yanfang Ye, and Chuxu Zhang. Co-modality graph contrastive learning for imbalanced node classification. In NeurIPS, volume 35, pages 15862–15874, 2022

    2022

  35. [35]

    Enhancing e-commerce representation learning via hypergraph contrastive learning and interpretable llm-driven analysis

    Yiyue Qian, Shinan Zhang, Lanhao Chen, Diego Socolinsky, Negin Sokhandan, Song Cui, De Chen, and Suchitra Sathyanarayana. Enhancing e-commerce representation learning via hypergraph contrastive learning and interpretable llm-driven analysis. In Companion WWW, pages 2512–2520, 2025

    2025

  36. [36]

    Collabeval: Enhancing llm-as-a-judge via multi-agent collaboration

    Yiyue Qian, Shinan Zhang, Yun Zhou, Haibo Ding, Diego Socolinsky, and Yi Zhang. Collabeval: Enhancing llm-as-a-judge via multi-agent collaboration. arXiv preprint arXiv:2603.00993, 2026

    work page arXiv 2026

  37. [37]

    Mali- cious repositories detection with adversarial heterogeneous graph contrastive learning

    Yiyue Qian, Yiming Zhang, Nitesh Chawla, Yanfang Ye, and Chuxu Zhang. Mali- cious repositories detection with adversarial heterogeneous graph contrastive learning. In CIKM, 2022

    2022

  38. [38]

    Rep2vec: Repository embedding via heterogeneous graph adversarial contrastive learning

    Yiyue Qian, Yiming Zhang, Qianlong Wen, Yanfang Ye, and Chuxu Zhang. Rep2vec: Repository embedding via heterogeneous graph adversarial contrastive learning. In KDD, 2022

    2022

  39. [39]

    Adapting meta knowl- edge with heterogeneous information network for covid-19 themed malicious repository detection

    Yiyue Qian, Yiming Zhang, Yanfang Ye, and Chuxu Zhang. Adapting meta knowl- edge with heterogeneous information network for covid-19 themed malicious repository detection. In IJCAI, 2021

    2021

  40. [40]

    Distilling meta knowledge on heterogeneous graph for illicit drug trafficker detection on social media

    Yiyue Qian, Yiming Zhang, Yanfang Ye, and Chuxu Zhang. Distilling meta knowledge on heterogeneous graph for illicit drug trafficker detection on social media. In NeurIPS, 2021

    2021

  41. [41]

    Learning to retrieve prompts for in-context learning

    Ohad Rubin, Jonathan Herzig, and Jonathan Berant. Learning to retrieve prompts for in-context learning. In NAACL-HLT, pages 2655–2671, 2022

    2022

  42. [42]

    A Systematic Survey of Prompt Engineering in Large Language Models: Techniques and Applications

    Pranab Sahoo, Ayush Kumar Singh, Sriparna Saha, Vinija Jain, Samrat Mondal, and Aman Chadha. A systematic survey of prompt engineering in large language models: Techniques and applications. arXiv preprint arXiv:2402.07927, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  43. [43]

    DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

    Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. Distilbert, a distilled version of bert: smaller, faster, cheaper and lighter. arXiv preprint arXiv:1910.01108, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1910

  44. [44]

    Learning from generalization patterns: An evaluation-driven approach to en- hanced data augmentation for fine-tuning small language models

    Huan Song, Deeksha Razdan, Yiyue Qian, Arijit Ghosh Chowdhury, Parth Patwa, Aman Chadha, Shinan Zhang, Sharlina Keshava, and Hannah Marlowe. Learning from generalization patterns: An evaluation-driven approach to en- hanced data augmentation for fine-tuning small language models. arXiv preprint arXiv:2510.18143, 2025

    work page arXiv 2025

  45. [45]

    An approach for the classification of urban building structures based on discriminant analysis techniques

    Stefan Steiniger, Tilman Lange, Dirk Burghardt, and Robert Weibel. An approach for the classification of urban building structures based on discriminant analysis techniques. Transactions in GIS, 12(1):31–59, 2008

    2008

  46. [46]

    Bim-to-brick: Using graph modeling for iot/bms and spatial semantic data interoperability within digital data models of buildings

    Filippo Vittori, Chuan Fu Tan, Anna Laura Pisello, Adrian Chong, Cristina Piselli, and Clayton Miller. Bim-to-brick: Using graph modeling for iot/bms and spatial semantic data interoperability within digital data models of buildings. Energy and Buildings, 348:116368, 2025

    2025

  47. [47]

    Building information modeling (bim) for existing buildings — literature review and future needs

    Roman Volk, Joachim Stengel, and Frank Schultmann. Building information modeling (bim) for existing buildings — literature review and future needs. Au- tomation in Construction, 38:109–127, 2014

    2014

  48. [48]

    To- wards evaluation guidelines for empirical studies involving llms

    Stefan Wagner, Marvin Muñoz Barón, Davide Falessi, and Sebastian Baltes. To- wards evaluation guidelines for empirical studies involving llms. arXiv preprint arXiv:2411.07668, 2024

    work page arXiv 2024

  49. [49]

    Automated point mapping for building control systems: Recent advances and future research needs

    Weimin Wang, Michael R Brambley, Woohyun Kim, Sriram Somasundaram, and Andrew J Stevens. Automated point mapping for building control systems: Recent advances and future research needs. Automation in Construction , 85:107–123, 2018

    2018

  50. [50]

    Emergent abilities of large language models

    Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, et al. Emergent abilities of large language models. Transactions on Machine Learning Research

  51. [51]

    Enhancing large language models through external domain knowledge

    Laslo Welz and Carsten Lanquillon. Enhancing large language models through external domain knowledge. In HCII, pages 135–146. Springer, 2024

    2024

  52. [52]

    Gcvr: reconstruction from cross-view enable sufficient and robust graph contrastive learning

    Qianlong Wen, Zhongyu Ouyang, Chunhui Zhang, Yiyue Qian, Chuxu Zhang, and Yanfang Ye. Gcvr: reconstruction from cross-view enable sufficient and robust graph contrastive learning. In UAI, 2024

    2024

  53. [53]

    Disentangled dynamic heterogeneous graph learning for opioid overdose prediction

    Qianlong Wen, Zhongyu Ouyang, Jianfei Zhang, Yiyue Qian, Yanfang Ye, and Chuxu Zhang. Disentangled dynamic heterogeneous graph learning for opioid overdose prediction. In KDD, 2022

    2022

  54. [54]

    Inference-time model steering via predictive-state intervention: A survey

    Renhao Xue, Rui Wang, Yawei Wang, Yueying Cui, Yiyue Qian, Praneetha Vad- damanu, Huan Song, and Hannah Marlowe. Inference-time model steering via predictive-state intervention: A survey. 2026

    2026

  55. [55]

    Domain knowledge is all you need: A field deployment of llm-powered test case generation in fintech domain

    Zhiyi Xue, Liangguo Li, Senyue Tian, Xiaohong Chen, Pingping Li, Liangyu Chen, Tingting Jiang, and Min Zhang. Domain knowledge is all you need: A field deployment of llm-powered test case generation in fintech domain. In ICSE-Companion, pages 314–315, 2024

    2024

  56. [56]

    Crossfit: A few-shot learning challenge for cross-task generalization in nlp

    Qinyuan Ye, Bill Yuchen Lin, and Xiang Ren. Crossfit: A few-shot learning challenge for cross-task generalization in nlp. In EMNLP, pages 7163–7189, 2021

    2021

  57. [57]

    Community mitigation: A data-driven system for covid-19 risk assessment in a hierarchical manner

    Yanfang Ye, Yujie Fan, Shifu Hou, Yiming Zhang, Yiyue Qian, Shiyu Sun, Qian Peng, Mingxuan Ju, Wei Song, and Kenneth Loparo. Community mitigation: A data-driven system for covid-19 risk assessment in a hierarchical manner. In CIKM, pages 2909–2916, 2020

    2020

  58. [58]

    𝛼-satellite: An ai-driven system and benchmark datasets for hierarchical community-level risk assessment to help combat covid-

    Yanfang Ye, Shifu Hou, Yujie Fan, Yiyue Qian, Yiming Zhang, Shiyu Sun, Qian Peng, and Kenneth Laparo. 𝛼-satellite: An ai-driven system and benchmark datasets for hierarchical community-level risk assessment to help combat covid-

  59. [59]

    arXiv preprint arXiv:2003.12232, 2020

    work page arXiv 2003

  60. [60]

    dstyle-gan: Generative adversarial network based on writing and photog- raphy styles for drug identification in darknet markets

    Yiming Zhang, Yiyue Qian, Yujie Fan, Yanfang Ye, Xin Li, Qi Xiong, and Fudong Shao. dstyle-gan: Generative adversarial network based on writing and photog- raphy styles for drug identification in darknet markets. In ACSAC, 2020

    2020

  61. [61]

    Adapting distilled knowledge for few-shot relation reasoning over knowledge graphs

    Yiming Zhang, Yiyue Qian, Yanfang Ye, and Chuxu Zhang. Adapting distilled knowledge for few-shot relation reasoning over knowledge graphs. In SDM, pages 666–674. SIAM, 2022

    2022

  62. [62]

    Calibrate before use: Improving few-shot performance of language models

    Zihao Zhao, Eric Wallace, Shi Feng, Dan Klein, and Sameer Singh. Calibrate before use: Improving few-shot performance of language models. In ICML, pages 12697–12706. PMLR, 2021

    2021