Publications

2025

1. 

   AI for Green Multi-Cluster: Intelligent Management towards Green and Low-Carbon, Large-scale Multi-Clusters

   Qizhen Weng, and Yuankai Fan

   In AI for Good Innovate for Impact Report Jul 2025

   Abs Bib HTML PDF

   This use case presents AI-powered green multi-cluster management for large-scale intelligent infrastructure. It uses AI-driven scheduling, GPU multiplexing, cross-cluster coordination, and multimodal explanation mechanisms to improve cluster utilization, reduce energy consumption, and support sustainable low-carbon AI infrastructure management.

   @incollection{weng2025GreenMultiCluster,
     author = {Weng, Qizhen and Fan, Yuankai},
     title = {AI for Green Multi-Cluster: Intelligent Management towards Green and Low-Carbon, Large-scale Multi-Clusters},
     booktitle = {AI for Good Innovate for Impact Report},
     publisher = {International Telecommunication Union},
     month = jul,
     year = {2025},
     chapter = {4.2-Climate Change},
     section = {Use Case 8},
     pages = {182--187},
   }

2. 

   Rethinking Data in NL2SQL: A Survey of What We Have and What We Expect

   Yuankai Fan, Qizhen Weng, Yin Chen, and X. Sean Wang

   Vicinagearth Nov 2025

   Abs Bib PDF

   This survey rethinks the role of data in NL2SQL, reviewing what data resources, benchmarks, and practices the community currently has and outlining what future NL2SQL systems and datasets should support.

   @article{fan2025NL2SQL,
     author = {Fan, Yuankai and Weng, Qizhen and Chen, Yin and Wang, X. Sean},
     title = {Rethinking Data in {NL2SQL}: A Survey of What We Have and What We Expect},
     journal = {Vicinagearth},
     volume = {2},
     number = {1},
     pages = {15},
     year = {2025},
     month = nov,
     issn = {3005-060X},
     doi = {10.1007/s44336-025-00026-9},
   }

3. 

   Grounding Natural Language to SQL Translation with Data-Based Self-Explanations

   Yuankai Fan, Tonghui Ren, Can Huang, Zhenying He, and X. Sean Wang

   In International Conference on Data Engineering 2025

   Abs Bib PDF Slides

   Natural Language Interfaces for Databases empower non-technical users to interact with data using natural language (NL). Advanced approaches, utilizing either neural sequence-to-sequence or more recent sophisticated large-scale language models, typically implement NL to SQL (NL2SQL) translation in an end-to-end fashion. However, like humans, these end-to-end translation models may not always generate the best SQL output on their first try. In this paper, we propose CycleSQL, an iterative framework designed for end-to-end translation models to autonomously generate the best output through self-evaluation. The main idea of CycleSQL is to introduce data-grounded NL explanations of query results as self-provided feedback, and use the feedback to validate the correctness of the translation iteratively, hence improving the overall translation accuracy.

   @inproceedings{cyclesql2025,
     author = {Fan, Yuankai and Ren, Tonghui and Huang, Can and He, Zhenying and Wang, X. Sean},
     title = {Grounding Natural Language to SQL Translation with Data-Based Self-Explanations},
     booktitle = {International Conference on Data Engineering},
     pages = {29--42},
     year = {2025},
   }

4. 

   The Power of Constraints in Natural Language to SQL Translation

   Tonghui Ren, Chen Ke,  Yuankai Fan, Yinan Jing, Zhenying He, Kai Zhang, and X. Sean Wang

   Proc. VLDB Endow. 2025

   Abs Bib PDF

   Current large language model (LLM)-based Natural Language to SQL (NL2SQL) approaches typically rely on the database schema and partial data values for the translation. These approaches are unable to use sufficient data for accurate database understanding due to limitations in data selection methods, and they cannot input the entire database due to the limited context window sizes of LLMs. This insufficient data integration may result in an incomplete understanding of the database, leading to semantically incorrect SQL generation. In this paper, we introduce REDSQL, a novel plug-and-play framework that refines the predicted SQL by utilizing the entire database in the refinement process.

   @article{red2025,
     author = {Ren, Tonghui and Ke, Chen and Fan, Yuankai and Jing, Yinan and He, Zhenying and Zhang, Kai and Wang, X. Sean},
     title = {The Power of Constraints in Natural Language to {SQL} Translation},
     journal = {Proc. {VLDB} Endow.},
     volume = {18},
     number = {7},
     pages = {2097--2111},
     year = {2025},
     doi = {10.14778/3734839.3734847},
   }

5. 

   Computation-Bandwidth-Memory Trade-offs: A Unified Paradigm for AI Infrastructure

   Yuankai Fan, Qizhen Weng, and Xuelong Li

   CoRR 2025

   Abs Bib PDF

   Large-scale artificial intelligence (AI) models are fundamentally transforming industries and redefining the paradigm of human–machine collaboration. While the technological revolution signals a new era of machine intelligence, the continued scaling of these models has exposed significant limitations in contemporary hardware architectures, manifesting as constraints on computational efficiency, interconnection bandwidth, and memory capacity. These three dimensions are inseparably intertwined, such that advances along any single axis often exacerbate bottlenecks in the others, rendering isolated optimizations increasingly ineffective. Achieving an optimal balance among them to maximize system efficiency therefore remains a central challenge in the design of scalable AI systems. To address this challenge, we introduce Computation-Bandwidth-Memory Trade-offs, termed the AI Trinity, a unified paradigm that positions computation, bandwidth, and memory as coequal pillars for next-generation AI infrastructure. At its core, AI Trinity enables a dynamic flow of resources across these pillars, transcending single-resource bottlenecks and adapting to diverse scenarios, thereby optimizing overall system performance to its maximum potential. Within this framework, AI Trinity identifies three fundamental trade-offs: (1) More Computation→Less Bandwidth, wherein computational resources are exploited to reduce data transmission under limited bandwidth conditions, (2) More Bandwidth→Less Memory, which exploits abundant communication capacity to populate or refresh memory when local storage resources are constrained, and (3) More Memory→Less Computation, whereby storage capacity are utilized to mitigate redundant computation when computational costs are prohibitive.

   @article{trinity2025,
     author = {Fan, Yuankai and Weng, Qizhen and Li, Xuelong},
     title = {Computation-Bandwidth-Memory Trade-offs: {A} Unified Paradigm for {AI} Infrastructure},
     journal = {CoRR},
     volume = {abs/2601.11577},
     year = {2025},
     doi = {10.48550/ARXIV.2601.11577},
     eprinttype = {arXiv},
     eprint = {2601.11577},
   }

2024

1. 

   A Confidence-based Knowledge Integration Framework for Cross-Domain Table Question Answering

   Yuankai Fan, Tonghui Ren, Can Huang, Beini Zheng, Yinan Jing, Zhenying He, Jinbao Li, and Jianxin Li

   Knowledge-Based Systems 2024

   Abs Bib PDF

   Recent advancements in TableQA leverage sequence-to-sequence (Seq2seq) deep learning models to accurately respond to natural language queries. These models achieve this by converting the queries into SQL queries, using information drawn from one or more tables. However, Seq2seq models often produce uncertain low-confidence predictions when distributing probability mass across multiple outputs during a decoding step, frequently yielding translation errors. To tackle this problem, we present CKIF, a confidence-based knowledge integration framework that uses a two-stage deep-learning-based ranking technique to mitigate the low-confidence problem commonly associated with Seq2seq models for TableQA.

   @article{ckif2024,
     author = {Fan, Yuankai and Ren, Tonghui and Huang, Can and Zheng, Beini and Jing, Yinan and He, Zhenying and Li, Jinbao and Li, Jianxin},
     title = {A Confidence-based Knowledge Integration Framework for Cross-Domain Table Question Answering},
     journal = {Knowledge-Based Systems},
     volume = {306},
     pages = {112718},
     year = {2024},
   }

2. 

   MetaSQL: A Generate-then-Rank Framework for Natural Language to SQL Translation

   Yuankai Fan, Zhenying He, Tonghui Ren, Can Huang, Yinan Jing, Kai Zhang, and X. Sean Wang

   In International Conference on Data Engineering 2024

   Abs Bib PDF Slides Video

   The Natural Language Interface to Databases (NLIDB) empowers non-technical users with database access through intuitive natural language interactions. Advanced approaches, utilizing neural sequence-to-sequence models or large-scale language models, typically employ auto-regressive decoding to generate unique SQL queries sequentially. In this paper, we propose MetaSQL, a unified generate-then-rank framework that can be flexibly incorporated with existing NLIDBs to consistently improve their translation accuracy. MetaSQL introduces query metadata to control the generation of better SQL query candidates and uses learning-to-rank algorithms to retrieve globally optimized queries.

   @inproceedings{metasql2024,
     author = {Fan, Yuankai and He, Zhenying and Ren, Tonghui and Huang, Can and Jing, Yinan and Zhang, Kai and Wang, X. Sean},
     title = {MetaSQL: A Generate-then-Rank Framework for Natural Language to SQL Translation},
     booktitle = {International Conference on Data Engineering},
     pages = {1765--1778},
     year = {2024},
   }

3. 

   PURPLE: Making a Large Language Model a Better SQL Writer

   Tonghui Ren,  Yuankai Fan, Zhenying He, Ren Huang, Jiaqi Dai, Can Huang, Yinan Jing, Kai Zhang, Yifan Yang, and X. Sean Wang

   In International Conference on Data Engineering 2024

   Abs Bib PDF

   Large Language Model (LLM) techniques play an increasingly important role in Natural Language to SQL (NL2SQL) translation. LLMs trained by extensive corpora have strong natural language understanding and basic SQL generation abilities without additional tuning specific to NL2SQL tasks. Existing LLMs-based NL2SQL approaches try to improve the translation by enhancing the LLMs with an emphasis on user intention understanding. However, LLMs sometimes fail to generate appropriate SQL due to their lack of knowledge in organizing complex logical operator composition. A promising method is to input the LLMs with demonstrations, which include known NL2SQL translations from various databases. LLMs can learn to organize operator compositions from the input demonstrations for the given task. In this paper, we propose PURPLE, which improves accuracy by retrieving demonstrations containing the requisite logical operator composition for the NL2SQL task on hand, thereby guiding LLMs to produce better SQL translation.

   @inproceedings{purple2024,
     author = {Ren, Tonghui and Fan, Yuankai and He, Zhenying and Huang, Ren and Dai, Jiaqi and Huang, Can and Jing, Yinan and Zhang, Kai and Yang, Yifan and Wang, X. Sean},
     title = {{PURPLE:} Making a Large Language Model a Better {SQL} Writer},
     booktitle = {International Conference on Data Engineering},
     pages = {15--28},
     publisher = {{IEEE}},
     year = {2024},
     doi = {10.1109/ICDE60146.2024.00009},
   }

4. 

   GAR++: Natural Language to SQL Translation with Efficient Generate-and-Rank

   Yuankai Fan, Can Huang, Tonghui Ren, Zhenying He, X. Sean Wang, Xianglian Wu, Yue Wang, Jiaming Li, and Yifan Yang

   In Asia-Pacific Web (APWeb) and Web-Age Information Management (WAIM) Joint International Conference on Web and Big Data 2024

   Abs Bib PDF

   Gar++ extends the existing generate-and-rank approach for a more efficient generation and robust ranking procedure.

   @inproceedings{gar++2024,
     author = {Fan, Yuankai and Huang, Can and Ren, Tonghui and He, Zhenying and Wang, X. Sean and Wu, Xianglian and Wang, Yue and Li, Jiaming and Yang, Yifan},
     title = {GAR++: Natural Language to {SQL} Translation with Efficient Generate-and-Rank},
     booktitle = {Asia-Pacific Web (APWeb) and Web-Age Information Management (WAIM) Joint International Conference on Web and Big Data},
     pages = {411--427},
     publisher = {Springer},
     year = {2024},
     doi = {10.1007/978-981-97-7238-4\_26},
   }

2023

1. 

   GAR: A Generate-and-Rank Approach for Natural Language to SQL Translation

   Yuankai Fan, Zhenying He, Tonghui Ren, Dianjun Guo, Lin Chen, Ruisi Zhu, Guanduo Chen, Yinan Jing, Kai Zhang, and X. Sean Wang

   In International Conference on Data Engineering 2023

   Abs Bib PDF Slides Video

   A Natural Language Interface to Databases (NLIDB) aims to help end-users access databases. State-of-the-art approaches primarily construct language translation models to convert NL queries to SQL queries. While these models exhibit good performance on NLIDB benchmarks, the translation accuracy seems to have stalled at between 70%-75%, and most erroneous translations happen with complex queries that require an understanding of the structure and semantics specific to a database. This paper proposes a Generate-And-Rank approach called GAR.

   @inproceedings{gar2023,
     author = {Fan, Yuankai and He, Zhenying and Ren, Tonghui and Guo, Dianjun and Chen, Lin and Zhu, Ruisi and Chen, Guanduo and Jing, Yinan and Zhang, Kai and Wang, X. Sean},
     title = {{GAR}: A Generate-and-Rank Approach for Natural Language to SQL Translation},
     booktitle = {International Conference on Data Engineering},
     pages = {110--122},
     year = {2023},
   }

2. 

   GenSQL: A Generative Natural Language Interface to Database Systems

   Yuankai Fan, Tonghui Ren, Zhenying He, X. Sean Wang, Ye Zhang, and Xingang Li

   In International Conference on Data Engineering Apr 2023

   Abs Bib PDF

   GenSql is a generative natural language interface to database systems that demonstrates natural language to SQL translation capabilities for interactive database access.

   @inproceedings{gensql2023,
     author = {Fan, Yuankai and Ren, Tonghui and He, Zhenying and Wang, X. Sean and Zhang, Ye and Li, Xingang},
     title = {{GenSQL}: A Generative Natural Language Interface to Database Systems},
     booktitle = {International Conference on Data Engineering},
     pages = {3603--3606},
     month = apr,
     year = {2023},
     doi = {10.1109/ICDE55515.2023.00278},
   }

3. 

   An Integrated Interactive Framework for Natural Language to SQL Translation

   Yuankai Fan, Tonghui Ren, Dianjun Guo, Zhigang Zhao, Zhenying He, X. Sean Wang, Yu Wang, and Tao Sui

   In International Conference on Web Information Systems Engineering 2023

   Abs Bib PDF

   An integrated interactive framework for natural language to SQL translation.

   @inproceedings{iknowsql2023,
     author = {Fan, Yuankai and Ren, Tonghui and Guo, Dianjun and Zhao, Zhigang and He, Zhenying and Wang, X. Sean and Wang, Yu and Sui, Tao},
     title = {An Integrated Interactive Framework for Natural Language to {SQL} Translation},
     booktitle = {International Conference on Web Information Systems Engineering},
     pages = {643--658},
     publisher = {Springer},
     year = {2023},
     doi = {10.1007/978-981-99-7254-8\_50},
   }

4. 

   Zebra: A novel method for optimizing text classification query in overload scenario

   Tianhuan Yu, Zhenying He, Zhihui Yang, Fei Ye,  Yuankai Fan, Yinan Jing, Kai Zhang, and X. Sean Wang

   World Wide Web Journal 2023

   Abs Bib PDF

   Zebra focuses on the query with text classification on streaming data. We propose a novel method called Zebra with progressive pipelines to optimize the overload query situations.

   @article{zebra2023,
     author = {Yu, Tianhuan and He, Zhenying and Yang, Zhihui and Ye, Fei and Fan, Yuankai and Jing, Yinan and Zhang, Kai and Wang, X. Sean},
     title = {Zebra: {A} novel method for optimizing text classification query in overload scenario},
     journal = {World Wide Web Journal},
     volume = {26},
     number = {3},
     pages = {905--931},
     year = {2023},
     doi = {10.1007/S11280-022-01061-Y},
   }

2022

1. 

   Staged query graph generation based on answer type for question answering over knowledge base

   Haoyuan Chen, Fei Ye,  Yuankai Fan, Zhenying He, Yinan Jing, Kai Zhang, and X. Sean Wang

   Knowl. Based Syst. 2022

   Abs Bib PDF

   Question answering over knowledge base (KBQA) enables users to query over the knowledge base without the need to know the details. A range of existing KBQA approaches treats the entities mentioned in the given question as the starting point to find the answers. While helpful in achieving improvements on the existing benchmarks, they have some limitations on the strategy of query graph generation, which creates too many candidate queries and makes it hard to select the best-matching one to get the answer. We propose a staged query graph generation approach based on the answer type, which exploits the correlation between questions and answer types to reduce the size of the candidate set and further improve the performance.

   @article{chen2022,
     author = {Chen, Haoyuan and Ye, Fei and Fan, Yuankai and He, Zhenying and Jing, Yinan and Zhang, Kai and Wang, X. Sean},
     title = {Staged query graph generation based on answer type for question answering over knowledge base},
     journal = {Knowl. Based Syst.},
     volume = {253},
     pages = {109576},
     year = {2022},
     doi = {10.1016/J.KNOSYS.2022.109576},
   }