pith. sign in

arxiv: 2606.29151 · v2 · pith:WYBVDUGMnew · submitted 2026-06-28 · 💻 cs.DB

CADENZA: Compiling Natural-Language Intent into Task-Specific Operator DAGs for Semantic Query Processing

Jaehyun Ha , Yongjoo Park , Wook-Shin Han This is my paper

Pith reviewed 2026-07-02 21:05 UTC · model grok-4.3

classification 💻 cs.DB
keywords semantic query processingtask DAGsrelational algebramulti-objective optimizationBayesian optimizationnatural language intentSQPE
0
0 comments X

The pith

CADENZA compiles each natural-language semantic operator intent into a space of typed task DAGs that can be rewritten and tuned for quality-latency-cost trade-offs.

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

Semantic query engines add model-based operators over unstructured data, but prior optimizers treat each operator as a black box and cannot filter, reorder, or jointly tune their intermediate outputs. CADENZA compiles every intent into an intent-specific plan space of typed task DAGs using a conservative extension of relational algebra called TxRA. The logical planner builds seed plans, applies dependency-checked structural rewrites, and enumerates semantics-guided alternatives; the physical planner routes tasks across backends and tunes cutpoints and thresholds via Bayesian optimization. This turns the multi-objective problem of quality, latency, and monetary cost into a searchable plan space. Experiments on SemBench report large gains in all three metrics over existing semantic query engines.

Core claim

CADENZA compiles each semantic operator instance—a template bound to a natural-language intent—into an intent-specific plan space of typed task DAGs and selects an executable plan under user-specified quality-latency-cost trade-offs. It does so by introducing task-extended relational algebra (TxRA), synthesizing seed TxRA plans, applying structural rewrites whose safety is checked from operator dependencies, enumerating semantics-guided alternatives, and jointly tuning routing cutpoints, backend parameters, and relational thresholds with Bayesian optimization.

What carries the argument

Task-extended relational algebra (TxRA), a conservative extension of relational algebra that treats task-specific operators and their intermediate outputs as first-class relational objects for filtering, reordering, routing, and thresholding.

If this is right

  • The logical planner can safely enumerate alternative task DAGs without changing the meaning of the original natural-language intent.
  • The physical planner can jointly optimize routing decisions, backend parameters, and relational thresholds under explicit quality-latency-cost constraints.
  • Intermediate task outputs become usable as relational objects, allowing standard algebraic identities to be applied to semantic operators.
  • Quality, latency, and cost can be traded off at the level of individual task instances rather than whole queries.

Where Pith is reading between the lines

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

  • The same compilation approach could be applied to other stochastic or model-driven operators that currently sit outside relational optimizers.
  • If the TxRA representation is exposed, existing relational engines might incorporate semantic operators without a full rewrite of their planners.
  • Users could receive multiple candidate DAGs ranked by different trade-off points rather than a single chosen plan.
  • The method might reduce reliance on hand-tuned prompts by turning intent into an explicit, rewritable plan space.

Load-bearing premise

Structural rewrites preserve query semantics when their safety conditions are verified only from operator dependencies.

What would settle it

A concrete query on SemBench where applying one of CADENZA’s structural rewrites produces a final answer whose semantic quality differs from the original intent under an independent equivalence check.

Figures

Figures reproduced from arXiv: 2606.29151 by Jaehyun Ha, Wook-Shin Han, Yongjoo Park.

Figure 1
Figure 1. Figure 1: A semantic operator instance embedded in SQL. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the CADENZA architecture illustrating an end-to-end workflow example with representative baseline plans. For clarity, implementation-level optimizations that apply uniformly across plans (e.g., LLM batching) are not shown. implementations (symbolic matching, a distilled QA model, a strong general-purpose LLM, and a composite RAG-style), using cheap input features such as text length with a tuna… view at source ↗
Figure 3
Figure 3. Figure 3: The workflow of CADENZA’s logical planner: seed synthesis (gray) and plan exploration (green). 𝜎answer=“yes”∧Score≥𝜆  ApplyTxtQA𝜑 (txt) [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: End-to-end per-query Q/L/C across systems for each scenario. [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Sensitivity to Bayesian optimization trials. [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: End-to-end scaling with data size. B Per-Scenario and BioDEX End-to-End Results Tables 7–11 report the per-scenario raw values summarized in [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 6
Figure 6. Figure 6: End-to-end scaling with data size. B Per-Scenario and BioDEX End-to-End Results Tables 7–11 report the per-scenario raw values summarized in [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Mean utility per (scenario, system, weight). Queries a system fails to run contribute zero to its mean (zero-fill). Higher [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
Figure 7
Figure 7. Figure 7: Mean utility per (scenario, system, weight). Queries a system fails to run contribute zero to its mean (zero-fill). Higher [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: BO trials and hypervolume over the Pareto set. [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
Figure 8
Figure 8. Figure 8: BO trials and hypervolume over the Pareto set. [PITH_FULL_IMAGE:figures/full_fig_p019_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Top-3 logical plans for a representative E-Commerce query. [PITH_FULL_IMAGE:figures/full_fig_p018_9.png] view at source ↗
Figure 9
Figure 9. Figure 9: Top-3 logical plans for a representative E-Commerce query. [PITH_FULL_IMAGE:figures/full_fig_p019_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Quality–Cost trade-off on Movie. Up-and-left is better. [PITH_FULL_IMAGE:figures/full_fig_p020_10.png] view at source ↗
Figure 10
Figure 10. Figure 10: Quality–Cost trade-off on Movie. Up-and-left is better. [PITH_FULL_IMAGE:figures/full_fig_p021_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Quality–Cost trade-off on Wildlife. Up-and-left is better. [PITH_FULL_IMAGE:figures/full_fig_p020_11.png] view at source ↗
Figure 11
Figure 11. Figure 11: Quality–Cost trade-off on Wildlife. Up-and-left is better. [PITH_FULL_IMAGE:figures/full_fig_p021_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Quality–Cost trade-off on MMQA. Up-and-left is better. [PITH_FULL_IMAGE:figures/full_fig_p021_12.png] view at source ↗
Figure 12
Figure 12. Figure 12: Quality–Cost trade-off on MMQA. Up-and-left is better. [PITH_FULL_IMAGE:figures/full_fig_p022_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Quality–Cost trade-off on Cars. Up-and-left is better. [PITH_FULL_IMAGE:figures/full_fig_p021_13.png] view at source ↗
Figure 13
Figure 13. Figure 13: Quality–Cost trade-off on Cars. Up-and-left is better. [PITH_FULL_IMAGE:figures/full_fig_p022_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Quality–Cost trade-off on E-Commerce. Up-and-left is better. [PITH_FULL_IMAGE:figures/full_fig_p022_14.png] view at source ↗
Figure 14
Figure 14. Figure 14: Quality–Cost trade-off on E-Commerce. Up-and-left is better. [PITH_FULL_IMAGE:figures/full_fig_p023_14.png] view at source ↗
read the original abstract

Semantic query processing engines (SQPEs) extend relational query processing with semantic operators that are executed via model inference over unstructured data. Optimizing such queries is inherently multi-objective: model inference dominates latency and monetary cost, and outputs are stochastic and backend-dependent, so quality must be optimized alongside efficiency. Existing SQPE optimizers do not expose each semantic operator instance's intermediate task outputs as a relational optimization object, leaving optimization unable to filter, reorder, route, threshold, or jointly tune them. We present CADENZA, which compiles each semantic operator instance--a template bound to a natural-language intent--into an intent-specific plan space of typed task DAGs and selects an executable plan under user-specified quality-latency-cost trade-offs. CADENZA introduces task-extended relational algebra (TxRA), a conservative extension of relational algebra with task-specific operators. The logical planner synthesizes seed TxRA plans, applies structural rewrites whose safety conditions are checked from operator dependencies, and enumerates semantics-guided alternatives from alternative-generation templates. The physical planner compiles each task-specific operator into a router over heterogeneous backends and jointly tunes routing cutpoints, backend parameters, and relational thresholds with Bayesian optimization. On SemBench, CADENZA improves the scenario-level averages of quality, latency, and cost by up to +0.49, 165.7x, and 310.3x, respectively, relative to state-of-the-art.

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

1 major / 0 minor

Summary. CADENZA compiles each semantic operator instance (a template bound to a natural-language intent) into an intent-specific plan space of typed task DAGs using task-extended relational algebra (TxRA), a conservative extension of relational algebra. The logical planner synthesizes seed TxRA plans, applies structural rewrites whose safety conditions are checked from operator dependencies, and enumerates semantics-guided alternatives; the physical planner compiles each task-specific operator into a router over heterogeneous backends and jointly tunes routing cutpoints, backend parameters, and relational thresholds via Bayesian optimization. On SemBench the system reports scenario-level average improvements of up to +0.49 in quality, 165.7x in latency, and 310.3x in cost relative to state-of-the-art SQPE optimizers.

Significance. If the central claims hold, the work would be a substantive contribution to semantic query processing by exposing intermediate task outputs as first-class relational optimization objects, enabling joint quality-latency-cost optimization that prior SQPE engines do not support. The conservative TxRA extension and the separation of dependency-checked logical rewriting from Bayesian physical tuning are technically interesting and could influence future designs of AI-augmented database systems.

major comments (1)
  1. [Abstract / Logical planner] Abstract / Logical planner description: the safety of structural rewrites is asserted on the basis of checks performed only from operator dependencies. Because semantic operators produce backend-dependent stochastic outputs, dependency-based checks alone do not establish that a rewritten DAG yields an output distribution whose quality is interchangeable with the original; any mismatch would invalidate both the quality gains and the subsequent physical tuning results. This assumption is load-bearing for the enumeration of alternatives and the reported improvements.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on our work. We respond to the major comment below, providing clarification on the safety conditions for structural rewrites in the logical planner.

read point-by-point responses

  1. Referee: [Abstract / Logical planner] Abstract / Logical planner description: the safety of structural rewrites is asserted on the basis of checks performed only from operator dependencies. Because semantic operators produce backend-dependent stochastic outputs, dependency-based checks alone do not establish that a rewritten DAG yields an output distribution whose quality is interchangeable with the original; any mismatch would invalidate both the quality gains and the subsequent physical tuning results. This assumption is load-bearing for the enumeration of alternatives and the reported improvements.

    Authors: The dependency checks verify that a structural rewrite preserves the exact dataflow: each task-specific operator receives identical inputs (including any upstream task outputs) as in the seed plan. Because each operator is a fixed function of its inputs, identical inputs imply identical output distributions regardless of later backend assignment. Backend dependence and stochasticity are addressed exclusively in the physical planner, which selects routers, tunes parameters, and sets thresholds via Bayesian optimization to achieve target quality. Consequently, enumerated alternatives remain valid at the task-semantics level, and quality measurements in the evaluation reflect these equivalent plans. We are prepared to add an explicit lemma formalizing input preservation (and thus distribution equivalence) under the checked conditions if the editor deems it necessary. revision: partial

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper introduces CADENZA, TxRA as a conservative extension of relational algebra, and describes logical/physical planners that synthesize plans, apply dependency-checked rewrites, and tune via Bayesian optimization. No equations, fitted parameters called predictions, self-citations, or uniqueness theorems appear in the abstract or described content that would reduce any claim to an input by construction. Performance numbers are presented as empirical results on the external SemBench benchmark, rendering the derivation self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 2 invented entities

Abstract only; no free parameters, axioms, or invented entities are detailed beyond high-level names.

invented entities (2)
  • TxRA no independent evidence
    purpose: Conservative extension of relational algebra with task-specific operators
    Introduced to expose intermediate task outputs for relational optimization
  • task-specific operator DAGs no independent evidence
    purpose: Represent intent-specific executable plans
    Compiled from natural-language intents

pith-pipeline@v0.9.1-grok · 5792 in / 1215 out tokens · 35108 ms · 2026-07-02T21:05:38.799834+00:00 · methodology

Review history (2 revisions) →

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

39 extracted references · 7 canonical work pages

  1. [1]

    Amazon Redshift ML

    2025. Amazon Redshift ML. https://aws.amazon.com/ko/redshift/features/ redshift-ml/. accessed: 2025-09-28

    2025

  2. [2]

    Samuel Arch, Yuchen Liu, Todd C Mowry, Jignesh M Patel, and Andrew Pavlo

  3. [3]

    The key to effective udf optimization: Before inlining, first perform outlin- ing.Proceedings of the VLDB Endowment18, 1 (2024), 1–13

    2024

  4. [4]

    Konstantinos Chasialis, Yannis Foufoulas, Alkis Simitsis, and Yannis Ioannidis

  5. [5]

    Optimizing UDF Queries in SQL Data Engines. (2025)

    2025

  6. [6]

    Karel D’Oosterlinck, François Remy, Johannes Deleu, Thomas Demeester, Chris Develder, Klim Zaporojets, Aneiss Ghodsi, Simon Ellershaw, Jack Collins, and Christopher Potts. 2023. BioDEX: Large-scale biomedical adverse drug event extraction for real-world pharmacovigilance. InFindings of the association for computational linguistics: EMNLP 2023. 13425–13454

    2023

  7. [7]

    David Eriksson and Martin Jankowiak. 2021. High-dimensional Bayesian opti- mization with sparse axis-aligned subspaces. InUncertainty in Artificial Intelli- gence. PMLR, 493–503

    2021

  8. [8]

    David Eriksson, Michael Pearce, Jacob Gardner, Ryan D Turner, and Matthias Poloczek. 2019. Scalable global optimization via local Bayesian optimization. Advances in neural information processing systems32 (2019)

    2019

  9. [9]

    Google. 2025. BigQuery ML. https://cloud.google.com/bigquery-ml/docs. [On- line; accessed 03-November-2025]

    2025

  10. [10]

    Sirui Hong, Yizhang Lin, Bang Liu, Bangbang Liu, Binhao Wu, Ceyao Zhang, Danyang Li, Jiaqi Chen, Jiayi Zhang, Jinlin Wang, et al. 2025. Data interpreter: An llm agent for data science. InFindings of the Association for Computational Linguistics: ACL 2025. 19796–19821

    2025

  11. [11]

    Shengran Hu, Cong Lu, and Jeff Clune. [n. d.]. Automated Design of Agentic Systems. InThe Thirteenth International Conference on Learning Representations

  12. [12]

    Fabian Hueske, Mathias Peters, Matthias J Sax, Astrid Rheinländer, Rico Bergmann, Aljoscha Krettek, and Kostas Tzoumas. 2012. Opening the Black Boxes in Data Flow Optimization.Proceedings of the VLDB Endowment5, 11 (2012)

    2012

  13. [13]

    Ilyas, Rahul Shah, Walid G

    Ihab F. Ilyas, Rahul Shah, Walid G. Aref, Jeffrey Scott Vitter, and Ahmed K. Elmagarmid. 2004. Rank-aware Query Optimization. InProceedings of the 2004 ACM SIGMOD International Conference on Management of Data (SIGMOD ’04). ACM, New York, NY, USA. doi:10.1145/1007568.1007593

    work page doi:10.1145/1007568.1007593 2004

  14. [14]

    Saehan Jo and Immanuel Trummer. 2024. Thalamusdb: Approximate query processing on multi-modal data.Proceedings of the ACM on Management of Data 2, 3 (2024), 1–26

    2024

  15. [15]

    Gaurav Tarlok Kakkar, Jiashen Cao, Pramod Chunduri, Zhuangdi Xu, Sury- atej Reddy Vyalla, Prashanth Dintyala, Anirudh Prabakaran, Jaeho Bang, Aubhro Sengupta, Kaushik Ravichandran, et al. 2023. Eva: An end-to-end exploratory video analytics system. InProceedings of the Seventh Workshop on Data Manage- ment for End-to-End Machine Learning. 1–5

    2023

  16. [16]

    Gaurav Tarlok Kakkar, Jiashen Cao, Aubhro Sengupta, Joy Arulraj, and Hyesoon Kim. 2025. Aero: Adaptive Query Processing of ML Queries.Proceedings of the ACM on Management of Data3, 3 (2025), 1–27

    2025

  17. [17]

    Tushar Khot, Harsh Trivedi, Matthew Finlayson, Yao Fu, Kyle Richardson, Peter Clark, and Ashish Sabharwal. [n. d.]. Decomposed Prompting: A Modular Ap- proach for Solving Complex Tasks. InThe Eleventh International Conference on Learning Representations

  18. [18]

    Jiale Lao, Andreas Zimmerer, Olga Ovcharenko, Tianji Cong, Matthew Russo, Gerardo Vitagliano, Michael Cochez, Fatma Özcan, Gautam Gupta, Thibaud Hottelier, et al. 2025. SemBench: A Benchmark for Semantic Query Processing Engines.arXiv preprint arXiv:2511.01716(2025)

    work page arXiv 2025

  19. [19]

    Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Veselin Stoyanov, and Luke Zettlemoyer. 2020. BART: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension. InProceedings of the 58th annual meeting of the association for computational linguistics. 7871–7880

    2020

  20. [20]

    Ilyas, and Sumin Song

    Chengkai Li, Kevin Chen-Chuan Chang, Ihab F. Ilyas, and Sumin Song. 2005. RankSQL: Query Algebra and Optimization for Relational Top- 𝑘 Queries. In Proceedings of the 2005 ACM SIGMOD International Conference on Management of Data (SIGMOD ’05). ACM, New York, NY, USA, 131–142. doi:10.1145/1066157. 1066173

    work page doi:10.1145/1066157 2005

  21. [21]

    Chunwei Liu, Matthew Russo, Michael Cafarella, Lei Cao, Peter Baile Chen, Zui Chen, Michael Franklin, Tim Kraska, Samuel Madden, Rana Shahout, et al. 2025. Palimpzest: Optimizing ai-powered analytics with declarative query processing. InProceedings of the Conference on Innovative Database Research (CIDR). 2

    2025

  22. [22]

    Jaehyun Nam, Jinsung Yoon, Jiefeng Chen, Jinwoo Shin, Sercan Ö Arık, and Tomas Pfister. 2025. MLE-STAR: Machine Learning Engineering Agent via Search and Targeted Refinement.arXiv preprint arXiv:2506.15692(2025)

    work page arXiv 2025

  23. [23]

    Liana Patel, Siddharth Jha, Melissa Pan, Harshit Gupta, Parth Asawa, Carlos Guestrin, and Matei Zaharia. 2025. Semantic Operators and Their Optimization: Enabling LLM-Based Data Processing with Accuracy Guarantees in LOTUS. Proceedings of the VLDB Endowment18, 11 (2025), 4171–4184

    2025

  24. [24]

    Archiki Prasad, Alexander Koller, Mareike Hartmann, Peter Clark, Ashish Sabhar- wal, Mohit Bansal, and Tushar Khot. 2024. Adapt: As-needed decomposition and planning with language models. InFindings of the Association for Computational Linguistics: NAACL 2024. 4226–4252

    2024

  25. [25]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. InInternational conference on machine learning. PmLR, 8748–8763

    2021

  26. [26]

    Matthew Russo, Sivaprasad Sudhir, Gerardo Vitagliano, Chunwei Liu, Tim Kraska, Samuel Madden, and Michael Cafarella. 2025. Abacus: A Cost-Based Optimizer for Semantic Operator Systems.arXiv preprint arXiv:2505.14661(2025)

    work page arXiv 2025

  27. [27]

    Jon Saad-Falcon, Adrian Gamarra Lafuente, Shlok Natarajan, Nahum Maru, Hristo Todorov, Etash Kumar Guha, E Kelly Buchanan, Mayee F Chen, Neel Guha, Christopher Re, et al. [n. d.]. An Architecture Search Framework for Inference- Time Techniques. InForty-second International Conference on Machine Learning

  28. [28]

    SemBench Maintainers. 2025. Medical scenario removed due to licensing issues. https://github.com/SemBench/SemBench/issues/16. GitHub Issue #16

    2025

  29. [29]

    Parameswaran, and Eugene Wu

    Shreya Shankar, Tristan Chambers, Tarak Shah, Aditya G. Parameswaran, and Eugene Wu. 2025. DocETL: Agentic Query Rewriting and Evaluation for Complex Document Processing.Proc. VLDB Endow.18, 9 (May 2025), 3035–3048. doi:10. 14778/3746405.3746426

    work page arXiv 2025

  30. [30]

    Matthias Urban and Carsten Binnig. 2024. CAESURA: language models as multi- modal query planners.CIDR(2024)

    2024

  31. [31]

    Jiayi Wang and Jianhua Feng. 2025. Unify: An unstructured data analytics system. In2025 IEEE 41st International Conference on Data Engineering (ICDE). IEEE Computer Society, 4662–4674

    2025

  32. [32]

    Junlin Wang, WANG Jue, Ben Athiwaratkun, Ce Zhang, and James Zou. [n. d.]. Mixture-of-Agents Enhances Large Language Model Capabilities. InThe Thir- teenth International Conference on Learning Representations

  33. [33]

    Jiayi Wang and Guoliang Li. 2025. Aop: Automated and interactive llm pipeline orchestration for answering complex queries. CIDR

    2025

  34. [34]

    Yongdong Wang, Runze Xiao, Jun Younes Louhi Kasahara, Ryosuke Yajima, Keiji Nagatani, Atsushi Yamashita, and Hajime Asama. 2024. Dart-llm: Dependency- aware multi-robot task decomposition and execution using large language models. arXiv preprint arXiv:2411.09022(2024)

    work page arXiv 2024

  35. [35]

    Yanbo Wang, Zixiang Xu, Yue Huang, Xiangqi Wang, Zirui Song, Lang Gao, Chenxi Wang, Xiangru Tang, Yue Zhao, Arman Cohan, et al . 2025. DyFlow: Dynamic Workflow Framework for Agentic Reasoning. InAdvances in Neural Information Processing Systems

    2025

  36. [36]

    Johannes Wehrstein, Tiemo Bang, Roman Heinrich, and Carsten Binnig. 2025. GRACEFUL: A Learned Cost Estimator for UDFs. In2025 IEEE 41st International CADENZA: Compiling Natural-Language Intent into Task-Specific Operator DAGs for Semantic Query Processing SIGMOD ’27, June 03–05, 2018, Woodstock, NY Conference on Data Engineering (ICDE). IEEE Computer Soci...

    2025

  37. [37]

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. React: Synergizing reasoning and acting in language models. InInternational Conference on Learning Representations (ICLR)

    2023

  38. [38]

    Enhao Zhang, Nicole Sullivan, Brandon Haynes, Ranjay Krishna, and Magdalena Balazinska. 2025. Self-Enhancing Video Data Management System for Com- positional Events with Large Language Models.Proceedings of the ACM on Management of Data3, 3 (2025), 1–29

    2025

  39. [39]

    #$%&”()"*+

    Jiayi Zhang, Jinyu Xiang, Zhaoyang Yu, Fengwei Teng, Xiong-Hui Chen, Jiaqi Chen, Mingchen Zhuge, Xin Cheng, Sirui Hong, Jinlin Wang, et al. [n. d.]. AFlow: Automating Agentic Workflow Generation. InThe Thirteenth International Con- ference on Learning Representations. SIGMOD ’27, June 03–05, 2018, Woodstock, NY Jaehyun Ha, Yongjoo Park, and Wook-Shin Han ...

    2018