Pith. sign in

REVIEW 3 major objections 6 minor 300 references

SCION is an auditable pipeline that induces schema graphs from raw text and, on a new 24-source benchmark, outperforms released schemas, classic ontology learning, and direct language-model baselines under all four schema-graph similarity m

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 13:31 UTC pith:2X3PNGUR

load-bearing objection SCOPE is a genuinely useful new benchmark for corpus-to-schema induction, and SCION is an honest, well-documented baseline; the main caveats—pretraining contamination and a few methodological details—are real but don't sink the core contribution. the 3 major comments →

arxiv 2607.21610 v1 pith:2X3PNGUR submitted 2026-05-20 cs.AI cs.LG

SCOPE and SCION: A Benchmark and an Auditable Reference Pipeline for Schema Induction and Fusion from Text

Miaobo Hu , Xiaobo Guo , Shuhao Hu , Bokun Wang , Rui Chen , Xin Wang , Daren Zha , Jun Xiao This is my paper
classification cs.AI cs.LG
keywords schema inductionontology learningbenchmarklarge language modelsschema fusioninformation extractionschema graphauditable pipeline
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper tries to establish that schema induction from raw text can be measured systematically, and that a pipeline which confines a large language model to a mined candidate space does it better than the alternatives. It introduces SCOPE, a 24-source benchmark where systems see only training text and are scored against gold schema graphs, and SCION, a reference pipeline with strict JSON contracts, evidence pointers, and deterministic fallbacks. On this benchmark, SCION-lite achieves the highest F1 among compared systems under all four schema-graph metrics. The authors are careful to frame the results as agreement with normalized typed-edge targets, not as proof that induced schemas surpass human ontology design.

Core claim

The paper's central claim is that corpus-to-schema induction and fusion can be benchmarked reproducibly, and that a candidate-constrained pipeline—SCION-lite—achieves the best F1 under all four schema-graph similarity metrics when compared with released source schemas, a Text2Onto-style baseline, LLM-only induction, and a matched extract-then-aggregate baseline. On the SCOPE core suite, SCION-lite records 0.7518 Literal, 0.9298 Fuzzy, 0.8909 Continuous, and 0.7888 Graph F1; a compact open-model variant, SCION-RL, reports even higher scores. The authors interpret these numbers as better agreement with the unified typed-edge target, not as a claim that induced schemas are intrinsically better

What carries the argument

The central mechanism is the candidate-space constraint with contract-constrained JSON generation. SCION first mines candidate entity, relation, event, and role items from train-only text, then lets the language model name, merge, and filter only within that candidate space. Every output must satisfy a strict JSON contract with evidence pointers, and deterministic validation with fallback removes non-compliant entries. This restriction prevents the model from freely inventing schema items and makes the whole induction process auditable through logs of retention, merging, and fallback.

Load-bearing premise

The load-bearing premise is that the language-model pipelines are truly inducing the schemas from the supplied training texts rather than recalling schemas they encountered during pretraining; the paper itself states that pretraining contamination cannot be completely ruled out even though its name-only probes show near-zero performance.

What would settle it

Construct a corpus paired with a newly written schema that is not present in any public language-model training data, run SCION-lite and the strongest direct extract-then-aggregate baseline under the SCOPE protocol on that corpus, and compare their scores with the margins reported in the paper; if the gap shrinks or reverses on uncontaminated schemas, the reported advantage is partly memorization rather than candidate-space reasoning.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Schema induction can be evaluated at the schema-graph level rather than only through downstream extraction, making the task measurable and reproducible.
  • Constraining a language model to a mined candidate space improves over direct LLM extraction and over classic mining-first ontology learning, suggesting the constraint itself is a source of gain.
  • The pipeline is cheap enough for routine use, averaging about four LLM calls and $0.12 per source, and a compact open-model variant reduces reliance on proprietary model providers.
  • Downstream fixed-extractor experiments show that the induced schema improves instance-level extraction F1 over released schemas and over the strongest direct baseline, with fusion adding a further gain.
  • Evidence-linked outputs and controllability statistics make schema induction auditable, supporting deployment in settings where provenance matters.

Where Pith is reading between the lines

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

  • If pretraining contamination is the only serious confound, a natural extension is to build a private or newly authored schema suite that could not appear in any language model's training data; the paper's own name-only probes suggest the risk is low, but such a test would settle it.
  • The candidate-space mechanism is broader than ontology induction: any structured generation task where outputs must be grounded in evidence could benefit from restricting the model to a pre-mined candidate set with deterministic validation.
  • The conservative fusion policy, which merges only when lexical, embedding, and structural checks agree and otherwise demotes items to extensions, could serve as a template for long-term schema maintenance under domain drift.
  • The core event-extraction target excludes inter-event links such as temporal and causal relations; extending the benchmark to include them would test whether the role-level gains persist when the schema graph becomes richer and more ambiguous.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper introduces SCOPE, a 24-source train-text-only benchmark for corpus-to-schema induction and optional fusion, built from 15 RE and 9 EE public datasets, with normalized gold schema graphs. It also presents SCION, a candidate-constrained, contract-LLM schema induction pipeline with optional conservative fusion, and a compact RL-trained variant (SCION-RL). The central empirical claim is that SCION-lite attains the highest F1 among the compared main-table systems under all four schema-graph similarity metrics (Literal, Fuzzy, Continuous, Graph) in Table 1, with reported gains over a matched extract-then-aggregate baseline (ETA). The paper includes extensive diagnostics: controllability logs, reachable-target evaluation, memorization probes, encoder sensitivity, human calibration, and downstream fixed-extractor experiments.

Significance. If the benchmark is valid and the results hold, SCOPE would fill a real gap: a reproducible, train-text-only evaluation protocol for schema induction and fusion, with gold graphs derived from public IE schemas and a clear separation of induction inputs from evaluation targets. The paper's strength is its auditability: code, run manifests, evidence-linked outputs, and detailed controllability statistics are promised in the repository. The comparison against released source schemas as a representation-formalism reference is thoughtful. The SCION pipeline itself is not a new extraction architecture, but as a reference baseline with constrained LLM generation it is useful. However, the benchmark's core measurement of 'induction' is threatened by possible LLM pretraining contamination, which the paper explicitly admits cannot be ruled out. The memorization probes are a good-faith attempt but are not decisive. The single-run main results and threshold tuning on the same benchmark further weaken the quantitative claims.

major comments (3)
  1. [Section 6 / Appendix Table 16] The pretraining-contamination concern is load-bearing for the benchmark's central claim of measuring corpus-to-schema induction, and the current evidence is not sufficient to rule it out. The name-only probes test a different input (schema names) than the real induction input (train texts from the same public datasets whose label vocabularies almost certainly appear in LLM pretraining corpora). The shuffled condition still yields C-F1 of 0.5047/0.5595, showing that lexical cues alone carry substantial signal. Since the paper's own Limitations state that 'pretraining contamination cannot be ruled out completely,' the absolute F1 scores and the gap over ETA may partly reflect retrieval rather than induction. To support the benchmark's validity, the authors should add a contamination-controlled analysis: e.g., a synthetic corpus with a schema invented after the models' knowledge cutoff, or
  2. [Section 5.3 / Appendix Table 7] The main Table 1 results are single-run point estimates, while run-to-run variance is reported only in the appendix as macro-averaged per-source standard deviations over 5 repeats on 8 sources. For Fuzzy and Continuous F1, the reported gains over ETA are +0.0157 and +0.0150, which are of the same order as the reported dispersion (SCION-lite std up to 0.015 for Continuous F1). Without paired significance tests, confidence intervals, or a per-source win-rate analysis, the claim that SCION-lite is 'highest' among main-table systems is not statistically supported. The authors should provide interval estimates for the macro-average differences or a per-source sign test across all 24 sources.
  3. [Section 5.3 / Appendix Table 6] The metric hyperparameters τ (Fuzzy threshold) and (α, K) (Graph-F1 smoothing) are selected on a calibration subset of SCOPE by maximizing macro F1. Although the calibration subset is fixed before scoring the reported systems, this is criterion-based tuning on the same benchmark that is then used for the main evaluation. This can inflate absolute scores and, more importantly, could differentially benefit methods that happen to produce label distributions aligned with the chosen threshold. The paper should report results for a default non-tuned setting (e.g., τ=0.5, α=0.5, K=1) and show that the method ranking in Table 1 is unchanged. The stability analysis in Appendix Table 13 addresses target formalism but not this hyperparameter-selection effect.
minor comments (6)
  1. [Section 5.7] The phrase 'highest F1 among the compared main-table systems' is technically correct, but since Appendix Table 10 shows SCION-RL outperforming SCION-lite, the text should explicitly state that SCION-RL is excluded from the main comparison because of its additional RL training stage, to avoid any impression of selective reporting.
  2. [Appendix Table 16] The 'name-only' and 'shuffled' conditions are not precisely defined in the text. Specify exactly what input is provided to the model in each condition (e.g., schema label lists with or without source names, versus shuffled text chunks) so that the probes can be interpreted and reproduced.
  3. [Table 7] The distinction between 'single-run (per source)' as the primary reporting mode and '5 repeats on 8 representative sources' for variance estimation should be stated in the main text (Section 5.3), not only in the appendix, to avoid misleading readers who see only Table 1.
  4. [Section 3.1] The paper uses 'ontology' and 'schema' loosely. It would help to define the scope of 'schema graph' versus 'ontology' early and consistently, particularly since the title and abstract use both terms.
  5. [Appendix A] The sentence 'The following prompts are experimental prompts used by the SCION induction modules; they are not instructions to reviewers or review tools' is unnecessary and out of place in a formal paper. Remove it.
  6. [Table 3] The note 'the value 48 is shown only for bookkeeping over configured module runs and is not the denominator of the reported rates' is confusing. Consider removing the 'Overall 48' column or explaining the arithmetic in the caption more clearly.

Circularity Check

0 steps flagged

No significant circularity: SCION is not fitted to gold and the main claim rests on SCION-lite; admitted contamination risk is a validity caveat, not a circular reduction.

full rationale

The paper is an empirical benchmark-and-pipeline contribution rather than a derivation, and the central claim (SCION-lite's Table 1 F1s) is produced from train-only induction texts and evaluated against split-independent gold schema graphs. The gold targets are never used in candidate mining, naming, merging, or SCION-RL training; the released-source-schema baseline is explicitly framed as a representation-formalism reference rather than an oracle. Metric hyperparameters (τ=0.45, α=0.5, K=2) are fixed on a calibration subset before scoring the reported systems and are not tuned per method, so no fitted threshold is renamed as a prediction. The use of the bge-m3 encoder both in the non-literal metrics and in SCION-full clustering is a shared-representation design choice that affects only the appendix ablation, not the main SCION-lite claim, and it does not make the metric equal to the pipeline output by construction. No load-bearing self-citations or imported uniqueness theorems appear. The admitted limitation that 'pretraining contamination cannot be ruled out completely' (Section 6, Limitations) is a validity threat to interpreting the F1s as induction rather than retrieval, but it is not a circularity reduction: the scores are not defined in terms of SCION's own outputs, and the memorization probes are an attempted independent check. Overall, no step in the claimed evaluation chain reduces to its own inputs.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

The central results rest on benchmark-conversion validity, embedding-metric validity, non-contamination of LLMs, and independence of the fusion base package. These are domain assumptions rather than standard math. No physical entities or new conserved quantities are invented; the ARG placeholder and SCOPE gold graphs are evaluation artifacts, not entities requiring independent falsifiable handles.

free parameters (4)
  • Fuzzy threshold tau = 0.45
    Selected by a validation sweep on a calibration subset of SCOPE to maximize macro Fuzzy F1 (Appendix Table 6); affects all Fuzzy F1 numbers in Table 1.
  • Graph smoothing alpha = 0.5
    Selected by a validation sweep on the same calibration subset to maximize macro Graph F1; chosen as a stable plateau value (Appendix Table 6).
  • Graph smoothing steps K = 2
    Selected by a validation sweep on the same calibration subset; chosen at the elbow of the Graph F1 curve (Appendix Table 6).
  • SCION-RL reward weights = (0.25, 0.20, 0.20, 0.10, 0.25)
    Hand-chosen weights over JSON validity, candidate linkage, evidence coverage, compactness, and structural consistency (Section 4.4); affects only the SCION-RL variant.
axioms (4)
  • domain assumption Gold schema graphs produced by converting each source's released schema to typed-edge form are valid and complete evaluation targets.
    Section 3.1 defines the gold graph as a deterministic conversion of released schemas. If normalization choices encode the authors' conventions, scores measure agreement with those conventions rather than intrinsic schema quality.
  • domain assumption Fuzzy/Continuous/Graph metrics using bge-m3 cosine similarity reflect meaningful semantic equivalence of schema labels.
    Appendix E defines the metrics; Appendix Table 18 shows human acceptance of only 0.2830 for high-score Fuzzy pairs, so this assumption is only partially supported.
  • domain assumption The LLM schema engineers are not materially contaminated by pretraining on the 24 public IE datasets.
    Section 6 states that pretraining contamination cannot be ruled out completely; Table 16's memorization probes reduce but do not eliminate the risk.
  • domain assumption The fixed base ontology package Obase is independent of the evaluation gold graphs.
    Section 5.5 and Appendix C.4 do not state how Obase was constructed. If it is derived from the same released schemas used to build gold graphs, fusion-track results would be partially circular.

pith-pipeline@v1.3.0-alltime-deepseek · 26881 in / 17407 out tokens · 203855 ms · 2026-08-02T13:31:30.496862+00:00 · methodology

0 comments
read the original abstract

Schema graphs are an upstream bottleneck of schema-grounded information extraction and knowledge graph construction, yet most extraction systems assume the schema is already available. We introduce SCOPE (Schema Construction and Ontology-induction Pipeline Evaluation), a train-text-only benchmark for corpus-to-schema induction and optional schema fusion from raw text, built from 24 public information extraction sources (15 RE and 9 EE) normalized into evaluation-only gold schema graphs; its core event-extraction target covers event types and within-event argument roles, with inter-event links reported separately. We present SCION (Schema Construction and Induction with Ontology Normalization), an auditable reference pipeline rather than a new extraction architecture; it constructs candidate spaces from train text and restricts naming, merging, filtering, validation, and conservative fusion to candidate-linked evidence under strict JSON contracts. On the SCOPE core suite, SCION-lite attains the highest F1 among released source-schema references, Text2Onto-style, LLM-only, and matched extract-then-aggregate baselines under Literal, Fuzzy, Continuous, and Graph schema-graph metrics, while the compact open-model SCION-RL variant reduces reliance on proprietary LLM schema engineers. These results are reported against normalized typed-edge targets rather than as claims that induced schemas surpass human ontology design; the release includes evidence-linked outputs, parse/fallback logs, candidate retention/merging logs, run manifests, code, and benchmark packages at https://github.com/wandugu/paper_scion.

Figures

Figures reproduced from arXiv: 2607.21610 by Bokun Wang, Daren Zha, Jun Xiao, Miaobo Hu, Rui Chen, Shuhao Hu, Xiaobo Guo, Xin Wang.

Figure 1
Figure 1. Figure 1: SCION framework overview. Starting from train-only text corpora and optional existing ontology packages, the system constructs a schema graph with relation types, event types, and within-event roles, then optionally fuses it into a unified schema/ontology artifact Ofusion for downstream knowledge graph extraction. to gold schema graphs; Appendix Tables 20 and 21 report zero-shot/SFT/RL and reward-term abla… view at source ↗

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

300 extracted references · 60 canonical work pages · 9 internal anchors

  1. [1]

    Instruction-

    Zhou, Jeffrey and Lu, Tianjian and Mishra, Swaroop and Brahma, Siddhartha and Basu, Sujoy and Luan, Yi and Zhou, Denny and Hou, Le , month = nov, year =. Instruction-. doi:10.48550/arXiv.2311.07911 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2311.07911

  2. [2]

    , month = nov, year =

    Rein, David and Hou, Betty Li and Stickland, Asa Cooper and Petty, Jackson and Pang, Richard Yuanzhe and Dirani, Julien and Michael, Julian and Bowman, Samuel R. , month = nov, year =. doi:10.48550/arXiv.2311.12022 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2311.12022

  3. [3]

    200多元带孩子吃了3个双人餐，纸包鱼铁锅炖香辣虾，没必要特别撑解馋就好\_哔哩哔哩\_bilibili , url =

    君在西安 , file =. 200多元带孩子吃了3个双人餐，纸包鱼铁锅炖香辣虾，没必要特别撑解馋就好\_哔哩哔哩\_bilibili , url =

  4. [4]

    doi:10.48550/arXiv.2310.16049 , abstract =

    Sprague, Zayne and Ye, Xi and Bostrom, Kaj and Chaudhuri, Swarat and Durrett, Greg , month = mar, year =. doi:10.48550/arXiv.2310.16049 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2310.16049

  5. [5]

    Challenging

    Suzgun, Mirac and Scales, Nathan and Schärli, Nathanael and Gehrmann, Sebastian and Tay, Yi and Chung, Hyung Won and Chowdhery, Aakanksha and Le, Quoc and Chi, Ed and Zhou, Denny and Wei, Jason , editor =. Challenging. Findings of the. 2023 , pages =. doi:10.18653/v1/2023.findings-acl.824 , abstract =

    doi:10.18653/v1/2023.findings-acl.824 2023

  6. [6]

    Proceedings of the 60th

    Lin, Stephanie and Hilton, Jacob and Evans, Owain , editor =. Proceedings of the 60th. 2022 , pages =. doi:10.18653/v1/2022.acl-long.229 , abstract =

    doi:10.18653/v1/2022.acl-long.229 2022

  7. [7]

    and Stoica, Ion , month = oct, year =

    Li, Tianle and Chiang, Wei-Lin and Frick, Evan and Dunlap, Lisa and Wu, Tianhao and Zhu, Banghua and Gonzalez, Joseph E. and Stoica, Ion , month = oct, year =. From. doi:10.48550/arXiv.2406.11939 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2406.11939

  8. [8]

    Zico and Fredrikson, Matt , month = dec, year =

    Zou, Andy and Wang, Zifan and Carlini, Nicholas and Nasr, Milad and Kolter, J. Zico and Fredrikson, Matt , month = dec, year =. Universal and. doi:10.48550/arXiv.2307.15043 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2307.15043

  9. [9]

    and Zhang, Hao and Gonzalez, Joseph E

    Zheng, Lianmin and Chiang, Wei-Lin and Sheng, Ying and Zhuang, Siyuan and Wu, Zhanghao and Zhuang, Yonghao and Lin, Zi and Li, Zhuohan and Li, Dacheng and Xing, Eric P. and Zhang, Hao and Gonzalez, Joseph E. and Stoica, Ion , month = dec, year =. Judging. doi:10.48550/arXiv.2306.05685 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2306.05685

  10. [10]

    , month = mar, year =

    Dubois, Yann and Galambosi, Balázs and Liang, Percy and Hashimoto, Tatsunori B. , month = mar, year =. Length-. doi:10.48550/arXiv.2404.04475 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2404.04475

  11. [11]

    Training

    Cobbe, Karl and Kosaraju, Vineet and Bavarian, Mohammad and Chen, Mark and Jun, Heewoo and Kaiser, Lukasz and Plappert, Matthias and Tworek, Jerry and Hilton, Jacob and Nakano, Reiichiro and Hesse, Christopher and Schulman, John , month = nov, year =. Training. doi:10.48550/arXiv.2110.14168 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2110.14168

  12. [12]

    Ganguli, Deep and Lovitt, Liane and Kernion, Jackson and Askell, Amanda and Bai, Yuntao and Kadavath, Saurav and Mann, Ben and Perez, Ethan and Schiefer, Nicholas and Ndousse, Kamal and Jones, Andy and Bowman, Sam and Chen, Anna and Conerly, Tom and DasSarma, Nova and Drain, Dawn and Elhage, Nelson and El-Showk, Sheer and Fort, Stanislav and Hatfield-Dodd...

    Pith/arXiv arXiv doi:10.48550/arxiv.2209.07858

  13. [13]

    Proceedings of the

    Völske, Michael and Potthast, Martin and Syed, Shahbaz and Stein, Benno , editor =. Proceedings of the. 2017 , pages =. doi:10.18653/v1/W17-4508 , abstract =

    doi:10.18653/v1/w17-4508 2017

  14. [14]

    and Daly, Raymond E

    Maas, Andrew L. and Daly, Raymond E. and Pham, Peter T. and Huang, Dan and Ng, Andrew Y. and Potts, Christopher , editor =. Learning. Proceedings of the 49th. 2011 , pages =

    2011

  15. [15]

    LIONs: An Empirically Optimized Approach to Align Language Models

    Yu, Xiao and Wu, Qingyang and Li, Yu and Yu, Zhou , month = oct, year =. doi:10.48550/arXiv.2407.06542 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2407.06542

  16. [16]

    Achieving

    Wang, Zhimeng and Wang, Pinzheng and Li, Juntao and Chen, Yibin and Zhang, Min , editor =. Achieving. Findings of the. 2024 , pages =. doi:10.18653/v1/2024.findings-emnlp.525 , abstract =

    work page doi:10.18653/v1/2024.findings-emnlp.525 2024

  17. [17]

    Break the

    Kim, Heegyu and Yuk, Sehyun and Cho, Hyunsouk , month = feb, year =. Break the. doi:10.48550/arXiv.2402.15180 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2402.15180

  18. [18]

    Understanding

    Ethayarajh, Kawin and Choi, Yejin and Swayamdipta, Swabha , month = jun, year =. Understanding. Proceedings of the 39th

  19. [19]

    Lightman, Hunter and Kosaraju, Vineet and Burda, Yura and Edwards, Harri and Baker, Bowen and Lee, Teddy and Leike, Jan and Schulman, John and Sutskever, Ilya and Cobbe, Karl , month = may, year =. Let's. doi:10.48550/arXiv.2305.20050 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2305.20050

  20. [20]

    Proceedings of the 2024

    Wang, Zhilin and Dong, Yi and Zeng, Jiaqi and Adams, Virginia and Sreedhar, Makesh Narsimhan and Egert, Daniel and Delalleau, Olivier and Scowcroft, Jane and Kant, Neel and Swope, Aidan and Kuchaiev, Oleksii , editor =. Proceedings of the 2024. 2024 , pages =. doi:10.18653/v1/2024.naacl-long.185 , abstract =

    doi:10.18653/v1/2024.naacl-long.185 2024

  21. [21]

    doi:10.48550/arXiv.2406.15513 , abstract =

    Ji, Jiaming and Hong, Donghai and Zhang, Borong and Chen, Boyuan and Dai, Josef and Zheng, Boren and Qiu, Tianyi and Li, Boxun and Yang, Yaodong , month = oct, year =. doi:10.48550/arXiv.2406.15513 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2406.15513

  22. [22]

    doi:10.48550/arXiv.2310.01377 , abstract =

    Cui, Ganqu and Yuan, Lifan and Ding, Ning and Yao, Guanming and He, Bingxiang and Zhu, Wei and Ni, Yuan and Xie, Guotong and Xie, Ruobing and Lin, Yankai and Liu, Zhiyuan and Sun, Maosong , month = jul, year =. doi:10.48550/arXiv.2310.01377 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2310.01377

  23. [23]

    doi:10.48550/arXiv.2501.16607 , abstract =

    Yuan, Shuozhi and Chen, Liming and Yuan, Miaomiao and Zhao, Jin and Peng, Haoran and Guo, Wenming , month = jan, year =. doi:10.48550/arXiv.2501.16607 , abstract =

    doi:10.48550/arxiv.2501.16607

  24. [24]

    doi:10.48550/arXiv.2312.11242 , abstract =

    Wang, Bing and Ren, Changyu and Yang, Jian and Liang, Xinnian and Bai, Jiaqi and Chai, Linzheng and Yan, Zhao and Zhang, Qian-Wen and Yin, Di and Sun, Xing and Li, Zhoujun , month = sep, year =. doi:10.48550/arXiv.2312.11242 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2312.11242

  25. [25]

    Hu, Miaobo and Wang, Bokun and Teng, Haoyuan and Yao, Hongyu and Zha, Daren and Xiao, Jun and Wang, Xin and Wang, Lei , keywords =

  26. [26]

    ISA Transactions , author =

    Cascade control of superheated steam temperature with neuro-. ISA Transactions , author =. 2012 , pages =. doi:10.1016/j.isatra.2012.06.008 , abstract =

    work page doi:10.1016/j.isatra.2012.06.008 2012

  27. [27]

    ISA Transactions , author =

    Cascade control method for hydraulic secondary regulation drive system based on adaptive robust control , volume =. ISA Transactions , author =. 2025 , pages =. doi:10.1016/j.isatra.2024.11.041 , abstract =

    work page doi:10.1016/j.isatra.2024.11.041 2025

  28. [28]

    ISA Transactions , author =

    Parallel cascade control of dead time processes via fractional order controllers based on. ISA Transactions , author =. 2020 , pages =. doi:10.1016/j.isatra.2019.08.047 , abstract =

    work page doi:10.1016/j.isatra.2019.08.047 2020

  29. [29]

    Procedia Computer Science , author =

    Particle. Procedia Computer Science , author =. 2016 , pages =. doi:10.1016/j.procs.2016.05.027 , abstract =

    work page doi:10.1016/j.procs.2016.05.027 2016

  30. [30]

    ISA Transactions , author =

    Phase leading active disturbance rejection control for a nanopositioning stage , volume =. ISA Transactions , author =. 2021 , pages =. doi:10.1016/j.isatra.2021.01.004 , abstract =

    work page doi:10.1016/j.isatra.2021.01.004 2021

  31. [31]

    Control Engineering Practice , author =

    Robust converter-fed motor control based on active rejection of multiple disturbances , volume =. Control Engineering Practice , author =. 2021 , pages =. doi:10.1016/j.conengprac.2020.104696 , abstract =

    arXiv 2021

  32. [32]

    Renewable and Sustainable Energy Reviews , author =

    Superheated steam drying:. Renewable and Sustainable Energy Reviews , author =. 2016 , pages =. doi:10.1016/j.rser.2016.03.033 , abstract =

    work page doi:10.1016/j.rser.2016.03.033 2016

  33. [33]

    Guan, Tongkun and Wang, Zining and Fu, Pei and Guo, Zhengtao and Shen, Wei and Zhou, Kai and Yue, Tiezhu and Duan, Chen and Sun, Hao and Jiang, Qianyi and Luo, Junfeng and Yang, Xiaokang , month = mar, year =. A. doi:10.48550/arXiv.2503.02304 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2503.02304

  34. [34]

    doi:10.48550/arXiv.2304.05302 , abstract =

    Yuan, Zheng and Yuan, Hongyi and Tan, Chuanqi and Wang, Wei and Huang, Songfang and Huang, Fei , month = oct, year =. doi:10.48550/arXiv.2304.05302 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2304.05302

  35. [35]

    Liu, Shunyu and Fang, Wenkai and Hu, Zetian and Zhang, Junjie and Zhou, Yang and Zhang, Kongcheng and Tu, Rongcheng and Lin, Ting-En and Huang, Fei and Song, Mingli and Li, Yongbin and Tao, Dacheng , month = mar, year =. A. doi:10.48550/arXiv.2503.11701 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2503.11701

  36. [36]

    Tihanyi, Norbert and Ferrag, Mohamed Amine and Jain, Ridhi and Bisztray, Tamas and Debbah, Merouane , month = sep, year =. 2024. doi:10.1109/CSR61664.2024.10679494 , abstract =

    arXiv 2024

  37. [37]

    doi:10.48550/arXiv.2401.08967 , abstract =

    Luong, Trung Quoc and Zhang, Xinbo and Jie, Zhanming and Sun, Peng and Jin, Xiaoran and Li, Hang , month = dec, year =. doi:10.48550/arXiv.2401.08967 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2401.08967

  38. [38]

    DeepSeek-AI and Guo, Daya and Yang, Dejian and Zhang, Haowei and Song, Junxiao and Zhang, Ruoyu and Xu, Runxin and Zhu, Qihao and Ma, Shirong and Wang, Peiyi and Bi, Xiao and Zhang, Xiaokang and Yu, Xingkai and Wu, Yu and Wu, Z. F. and Gou, Zhibin and Shao, Zhihong and Li, Zhuoshu and Gao, Ziyi and Liu, Aixin and Xue, Bing and Wang, Bingxuan and Wu, Bocha...

    Pith/arXiv arXiv doi:10.48550/arxiv.2501.12948

  39. [39]

    Ouyang, Long and Wu, Jeff and Jiang, Xu and Almeida, Diogo and Wainwright, Carroll L. and Mishkin, Pamela and Zhang, Chong and Agarwal, Sandhini and Slama, Katarina and Ray, Alex and Schulman, John and Hilton, Jacob and Kelton, Fraser and Miller, Luke and Simens, Maddie and Askell, Amanda and Welinder, Peter and Christiano, Paul and Leike, Jan and Lowe, R...

    Pith/arXiv arXiv doi:10.48550/arxiv.2203.02155

  40. [40]

    Advances in Neural Information Processing Systems , author =

    Qlora:. Advances in Neural Information Processing Systems , author =. 2024 , keywords =

    2024

  41. [41]

    Yan, Yuzi and Lou, Xingzhou and Li, Jialian and Zhang, Yiping and Xie, Jian and Yu, Chao and Wang, Yu and Yan, Dong and Shen, Yuan , month = oct, year =. Reward-. doi:10.48550/arXiv.2409.15360 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2409.15360

  42. [42]

    Proximal

    Schulman, John and Wolski, Filip and Dhariwal, Prafulla and Radford, Alec and Klimov, Oleg , month = aug, year =. Proximal. doi:10.48550/arXiv.1707.06347 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.1707.06347

  43. [43]

    Rusu, Andrei A. and Colmenarejo, Sergio Gomez and Gulcehre, Caglar and Desjardins, Guillaume and Kirkpatrick, James and Pascanu, Razvan and Mnih, Volodymyr and Kavukcuoglu, Koray and Hadsell, Raia , month = jan, year =. Policy

  44. [44]

    Mnih, Volodymyr and Kavukcuoglu, Koray and Silver, David and Graves, Alex and Antonoglou, Ioannis and Wierstra, Daan and Riedmiller, Martin , month = dec, year =. Playing

  45. [45]

    Mixtures of

    Obando-Ceron, Johan and Sokar, Ghada and Willi, Timon and Lyle, Clare and Farebrother, Jesse and Foerster, Jakob and Dziugaite, Gintare Karolina and Precup, Doina and Castro, Pablo Samuel , month = jun, year =. Mixtures of. doi:10.48550/arXiv.2402.08609 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2402.08609

  46. [46]

    Nature , author =

    Mastering the game of. Nature , author =. 2016 , pages =. doi:10.1038/nature16961 , language =

    doi:10.1038/nature16961 2016

  47. [47]

    Mastering

    Silver, David and Hubert, Thomas and Schrittwieser, Julian and Antonoglou, Ioannis and Lai, Matthew and Guez, Arthur and Lanctot, Marc and Sifre, Laurent and Kumaran, Dharshan and Graepel, Thore and Lillicrap, Timothy and Simonyan, Karen and Hassabis, Demis , month = dec, year =. Mastering. doi:10.48550/arXiv.1712.01815 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.1712.01815

  48. [48]

    and Tegmark, Max , month = may, year =

    Liu, Ziming and Wang, Yixuan and Vaidya, Sachin and Ruehle, Fabian and Halverson, James and Soljačić, Marin and Hou, Thomas Y. and Tegmark, Max , month = may, year =

  49. [49]

    Lepikhin, Dmitry and Lee, HyoukJoong and Xu, Yuanzhong and Chen, Dehao and Firat, Orhan and Huang, Yanping and Krikun, Maxim and Shazeer, Noam and Chen, Zhifeng , month = jun, year =

  50. [50]

    and Harley, Tim and Silver, David and Kavukcuoglu, Koray , month = jun, year =

    Mnih, Volodymyr and Badia, Adrià Puigdomènech and Mirza, Mehdi and Graves, Alex and Lillicrap, Timothy P. and Harley, Tim and Silver, David and Kavukcuoglu, Koray , month = jun, year =. Asynchronous

  51. [51]

    Reed, Scott and Zolna, Konrad and Parisotto, Emilio and Colmenarejo, Sergio Gomez and Novikov, Alexander and Barth-Maron, Gabriel and Gimenez, Mai and Sulsky, Yury and Kay, Jackie and Springenberg, Jost Tobias and Eccles, Tom and Bruce, Jake and Razavi, Ali and Edwards, Ashley and Heess, Nicolas and Chen, Yutian and Hadsell, Raia and Vinyals, Oriol and Bo...

    Pith/arXiv arXiv doi:10.48550/arxiv.2205.06175

  52. [52]

    Chen, Mark and Tworek, Jerry and Jun, Heewoo and Yuan, Qiming and Pinto, Henrique Ponde de Oliveira and Kaplan, Jared and Edwards, Harri and Burda, Yuri and Joseph, Nicholas and Brockman, Greg and Ray, Alex and Puri, Raul and Krueger, Gretchen and Petrov, Michael and Khlaaf, Heidy and Sastry, Girish and Mishkin, Pamela and Chan, Brooke and Gray, Scott and...

    Pith/arXiv arXiv doi:10.48550/arxiv.2107.03374

  53. [53]

    Entropy , author =

    Chess. Entropy , author =. 2022 , note =. doi:10.3390/e24040550 , abstract =

    doi:10.3390/e24040550 2022

  54. [54]

    Instruction

    Peng, Baolin and Li, Chunyuan and He, Pengcheng and Galley, Michel and Gao, Jianfeng , month = apr, year =. Instruction. doi:10.48550/arXiv.2304.03277 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2304.03277

  55. [55]

    Alpaca: A Strong, Replicable Instruction-Following Model , author =

    Alpaca:. Alpaca: A Strong, Replicable Instruction-Following Model , author =

  56. [56]

    Workshop, BigScience and Scao, Teven Le and Fan, Angela and Akiki, Christopher and Pavlick, Ellie and Ilić, Suzana and Hesslow, Daniel and Castagné, Roman and Luccioni, Alexandra Sasha and Yvon, François and Gallé, Matthias and Tow, Jonathan and Rush, Alexander M. and Biderman, Stella and Webson, Albert and Ammanamanchi, Pawan Sasanka and Wang, Thomas and...

    Pith/arXiv arXiv doi:10.48550/arxiv.2211.05100

  57. [57]

    Agent57:

    Badia, Adrià Puigdomènech and Piot, Bilal and Kapturowski, Steven and Sprechmann, Pablo and Vitvitskyi, Alex and Guo, Daniel and Blundell, Charles , month = mar, year =. Agent57:. doi:10.48550/arXiv.2003.13350 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2003.13350 2003

  58. [58]

    Huang, Wenlong and Xia, Fei and Xiao, Ted and Chan, Harris and Liang, Jacky and Florence, Pete and Zeng, Andy and Tompson, Jonathan and Mordatch, Igor and Chebotar, Yevgen and Sermanet, Pierre and Brown, Noah and Jackson, Tomas and Luu, Linda and Levine, Sergey and Hausman, Karol and Ichter, Brian , month = jul, year =. Inner. doi:10.48550/arXiv.2207.0560...

    Pith/arXiv arXiv doi:10.48550/arxiv.2207.05608

  59. [59]

    and Martic, Miljan and Legg, Shane and Amodei, Dario , month = feb, year =

    Christiano, Paul and Leike, Jan and Brown, Tom B. and Martic, Miljan and Legg, Shane and Amodei, Dario , month = feb, year =. Deep reinforcement learning from human preferences , url =. doi:10.48550/arXiv.1706.03741 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.1706.03741

  60. [60]

    Layer-wise

    Binder, Alexander and Montavon, Grégoire and Bach, Sebastian and Müller, Klaus-Robert and Samek, Wojciech , month = apr, year =. Layer-wise. doi:10.48550/arXiv.1604.00825 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.1604.00825

  61. [61]

    Flamingo: a

    Alayrac, Jean-Baptiste and Donahue, Jeff and Luc, Pauline and Miech, Antoine and Barr, Iain and Hasson, Yana and Lenc, Karel and Mensch, Arthur and Millican, Katie and Reynolds, Malcolm and Ring, Roman and Rutherford, Eliza and Cabi, Serkan and Han, Tengda and Gong, Zhitao and Samangooei, Sina and Monteiro, Marianne and Menick, Jacob and Borgeaud, Sebasti...

    Pith/arXiv arXiv doi:10.48550/arxiv.2204.14198

  62. [62]

    Smith, Shaden and Patwary, Mostofa and Norick, Brandon and LeGresley, Patrick and Rajbhandari, Samyam and Casper, Jared and Liu, Zhun and Prabhumoye, Shrimai and Zerveas, George and Korthikanti, Vijay and Zhang, Elton and Child, Rewon and Aminabadi, Reza Yazdani and Bernauer, Julie and Song, Xia and Shoeybi, Mohammad and He, Yuxiong and Houston, Michael a...

    Pith/arXiv arXiv doi:10.48550/arxiv.2201.11990

  63. [63]

    and Finn, Chelsea , month = jul, year =

    Rafailov, Rafael and Sharma, Archit and Mitchell, Eric and Ermon, Stefano and Manning, Christopher D. and Finn, Chelsea , month = jul, year =. Direct. doi:10.48550/arXiv.2305.18290 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2305.18290

  64. [64]

    Kolmogorov-

    Xu, Kunpeng and Chen, Lifei and Wang, Shengrui , month = jun, year =. Kolmogorov-. doi:10.48550/arXiv.2406.02496 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2406.02496

  65. [65]

    Puigcerver, Joan and Riquelme, Carlos and Mustafa, Basil and Houlsby, Neil , month = may, year =. From. doi:10.48550/arXiv.2308.00951 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2308.00951

  66. [66]

    Learning from delayed rewards -

  67. [67]

    , year =

    Bellman, Richard E. , year =. Dynamic

  68. [68]

    Dynamic programming , isbn =

    Bellman, Richard , year =. Dynamic programming , isbn =

  69. [69]

    Janner, Michael and Li, Qiyang and Levine, Sergey , month = nov, year =. Offline. doi:10.48550/arXiv.2106.02039 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2106.02039

  70. [70]

    and Broshear, Jackson and Schrittwieser, Julian and Hubert, Thomas and Vinyals, Oriol and Mann, Timothy , month = feb, year =

    Mandhane, Amol and Zhernov, Anton and Rauh, Maribeth and Gu, Chenjie and Wang, Miaosen and Xue, Flora and Shang, Wendy and Pang, Derek and Claus, Rene and Chiang, Ching-Han and Chen, Cheng and Han, Jingning and Chen, Angie and Mankowitz, Daniel J. and Broshear, Jackson and Schrittwieser, Julian and Hubert, Thomas and Vinyals, Oriol and Mann, Timothy , mon...

    Pith/arXiv arXiv doi:10.48550/arxiv.2202.06626

  71. [71]

    Nature , author =

    Mastering the game of. Nature , author =. 2017 , pages =. doi:10.1038/nature24270 , language =

    doi:10.1038/nature24270 2017

  72. [72]

    Weng, Lilian , month = apr, year =. Policy

  73. [73]

    Ji, Tianrui and Hou, Yuntian and Zhang, Di , month = dec, year =. A. doi:10.48550/arXiv.2407.11075 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2407.11075

  74. [74]

    Kaufmann, Timo and Weng, Paul and Bengs, Viktor and Hüllermeier, Eyke , month = apr, year =. A. doi:10.48550/arXiv.2312.14925 , abstract =

    doi:10.48550/arxiv.2312.14925

  75. [75]

    Training

    Black, Kevin and Janner, Michael and Du, Yilun and Kostrikov, Ilya and Levine, Sergey , month = jan, year =. Training. doi:10.48550/arXiv.2305.13301 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2305.13301

  76. [76]

    and Vinyals, Oriol and Sifre, Laurent , month = mar, year =

    Hoffmann, Jordan and Borgeaud, Sebastian and Mensch, Arthur and Buchatskaya, Elena and Cai, Trevor and Rutherford, Eliza and Casas, Diego de Las and Hendricks, Lisa Anne and Welbl, Johannes and Clark, Aidan and Hennigan, Tom and Noland, Eric and Millican, Katie and Driessche, George van den and Damoc, Bogdan and Guy, Aurelia and Osindero, Simon and Simony...

    Pith/arXiv arXiv doi:10.48550/arxiv.2203.15556

  77. [77]

    Decision

    Chen, Lili and Lu, Kevin and Rajeswaran, Aravind and Lee, Kimin and Grover, Aditya and Laskin, Michael and Abbeel, Pieter and Srinivas, Aravind and Mordatch, Igor , month = jun, year =. Decision. doi:10.48550/arXiv.2106.01345 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2106.01345

  78. [78]

    Outrageously

    Shazeer, Noam and Mirhoseini, Azalia and Maziarz, Krzysztof and Davis, Andy and Le, Quoc and Hinton, Geoffrey and Dean, Jeff , month = jan, year =. Outrageously. doi:10.48550/arXiv.1701.06538 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.1701.06538

  79. [79]

    2025 , file =

    Maxwell-. 2025 , file =

    2025

  80. [80]

    doi:10.48550/arXiv.2403.07974 , abstract =

    Jain, Naman and Han, King and Gu, Alex and Li, Wen-Ding and Yan, Fanjia and Zhang, Tianjun and Wang, Sida and Solar-Lezama, Armando and Sen, Koushik and Stoica, Ion , month = jun, year =. doi:10.48550/arXiv.2403.07974 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2403.07974

Showing first 80 references.