Efficient and Controllable Model Compression through Sequential Knowledge Distillation and Pruning
Open accessThe DOI matches its Crossref record; 21 of the 21 references it cites were found in a registry or an index; 14 of the 19 sources its claims cite were checked for support.
Abstract
Efficient model deployment is a key focus in deep learning. This has led to the exploration of methods such as knowledge distillation and network pruning to compress models and increase their performance. In this study, we investigate the potential synergy between knowledge distillation and network pruning to achieve optimal model efficiency and improved generalization. We introduce an innovative framework for model compression that combines knowledge distillation, pruning, and fine-tuning to achieve enhanced compression while providing control over the degree of compactness. Our research is conducted on popular datasets, CIFAR-10 and CIFAR-100, employing diverse model architectures, including ResNet, DenseNet, and EfficientNet. We could calibrate the amount of compression achieved. This allows us to produce models with different degrees of compression while still being just as accurate, or even better. Notably, we demonstrate its efficacy by producing two compressed variants of ResNet 101: ResNet 50 and ResNet 18. Our results reveal intriguing findings. In most cases, the pruned and distilled student models exhibit comparable or superior accuracy to the distilled student models while utilizing significantly fewer parameters.
Overview
CoChat Summary
TL;DR
The paper proposes a sequential compression pipeline that applies SimKD knowledge distillation first, then L1 unstructured pruning, then fine-tuning only the last fully connected layer. Across CIFAR-10 and CIFAR-100 with 11 teacher-student model combinations, it reports substantial parameter reduction while often matching or exceeding teacher accuracy.
Key findings
- SimKD consistently outperformed the compared distillation methods on CIFAR-100 and CIFAR-10, and its distilled students were selected for pruning.
- Pruned student models often achieved comparable or higher accuracy than teachers with far fewer parameters; the paper reports more than 70% compression of the teacher model across 11 model combinations, with some cases over 90%.
- For ResNet101→ResNet50 on CIFAR-100, the pruned student reached 58.92% accuracy with 0.5 × 10^7 parameters versus the teacher’s 54.07% with 4.45 × 10^7 parameters, an 88.76% parameter reduction. On CIFAR-10, the same pair reached 83.51% with 0.6 × 10^7 parameters versus the teacher’s 81.1% with 4.45 × 10^7 parameters, an 86.51% reduction.
- The paper reports two compressed ResNet101 variants: one pruned from ResNet50 with a 7.41× compression rate and 83.5% accuracy, and one pruned from ResNet18 with a 13.08× compression rate and 80.78% accuracy, about 1% lower than its original version.
Methods
- Pipeline: knowledge distillation using SimKD with teacher classifier reuse and feature alignment, followed by L1 unstructured pruning and fine-tuning of the last fully connected layer.
- Data and models: CIFAR-10 and CIFAR-100, with ResNet, DenseNet, EfficientNet, and MobileNet architectures across 11 teacher-student combinations; models were pretrained on ImageNet.
- Training and analysis: SGD with 0.9 Nesterov momentum, 50 epochs, learning rate 0.001, batch size 64, weight decay 5 × 10^-4, temperature T = 4, PyTorch 2.0.0 on a Quadro RTX 8000. Pruning reduced parameters from 0% to 90% in 10% increments, repeated 10 times per teacher-student pair and averaged; accuracy and remaining parameters were compared, including a “breakpoint” threshold.
Limitations
- Stated limitations: The paper does not include a dedicated limitations section; it instead lists future work such as combining KD with structured or channel pruning, optimizing fine-tuning, and evaluating on wider and more challenging datasets.
- Evident from the text: Evaluation is limited to CIFAR-10 and CIFAR-100 after ImageNet pretraining, and fine-tuning is applied only to the last fully connected layer rather than the whole network.
- Evident from the text: The pruning experiments are conducted on SimKD-distilled student models, so the reported pruning results are tied to that specific distillation choice.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
5 Valid · 1 Not valid · 4 Not checked · 42 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“Knowledge distillation provides a degree of controllable compression by allowing us to transfer knowledge from a larger model to a smaller one.”
- Also stated in: § 2.2. Pruning · plain
“This is where Knowledge Distillation (KD) comes in and offers a solution by introducing a semi-controlled form of compression.”
What the sources say
- Supportedcites [1] Hinton, G.E., Vinyals, O., and Dean, J. (2015). Distilling the Knowledge in a Neural Network. arXiv. source
The Introduction (p. 1) defines distillation as transferring knowledge from a cumbersome model to a smaller model suited to deployment. The Discussion (p. 8) reports successful transfer from an ensemble or large model into a smaller distilled model, supporting the claim's description of model compression. (LLM, conf=0.90)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 2.2. Pruning · plain
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Our approach provides a clear advantage by enabling a greater compression of the teacher model compared to the previous methods that relied on pruning [10-15], knowledge distillation (KD) [1,3,5,7,8], and the combination of both [16-20].”
What the sources say
- Supportedcites [10] Han, S., Mao, H., and Dally, W.J. (2016). Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding. arXiv. source
The citation is attached to prior methods that relied on pruning, not to the citing authors’ comparison with their own approach. The Introduction (p. 2) identifies pruning as the first stage of this paper’s compression method, and Section 2 (p. 2) describes how it removes connections to reduce model size. (LLM, conf=0.91)
- Supportedcites [11] Molchanov, P., Mallya, A., Tyree, S., Frosio, I., and Kautz, J. (2019, January 15–20). Importance Estimation for Neural Network Pruning. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, CA, USA. source
The citation is attached to “previous methods that relied on pruning,” not to the citing authors’ claim that their own approach achieves greater compression. The Abstract (p. 1) and Introduction (p. 1) identify this source as a method that prunes trained neural networks by removing less-important filters. (LLM, conf=0.92)
- Supportedcites [12] Zhou, G., Fan, Y., Cui, R., Bian, W., Zhu, X., and Gai, K. (2018, January 2–7). Rocket Launching: A Universal and Efficient Framework for Training Well-Performing Light Net. Proceedings of the AAAI Conference on Artificial Intelligence, New Orleans, LA, USA. source
The citation identifies a prior teacher–student model-compression method, not the source of the citing authors’ claim that their own approach achieves greater compression. The Introduction (p. 1) describes Rocket Launching as a framework for training small networks with a more complex network’s help, and “Characters of our method” (p. 3) explicitly considers knowledge distillation as a hint-loss function. (LLM, conf=0.85)
- Supportedcites [13] Ding, X., Ding, G., Guo, Y., Han, J., and Yan, C. (2019, January 10–15). Approximated Oracle Filter Pruning for Destructive CNN Width Optimization 2019. Proceedings of the 36th International Conference on Machine Learning, Long Beach, CA, USA. source
The citation is attached to “previous methods that relied on pruning,” not to the citing authors’ claim that their own approach achieves greater compression. The cited paper’s Introduction (p. 1) explicitly categorizes its method as filter pruning, so it supports that identification. (LLM, conf=0.92)
- Supportedcites [14] Frankle, J., and Carbin, M. (2019, January 9–15). The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks. Proceedings of the International Conference on Machine Learning, Long Beach, CA, USA. source
The citation is attached to “previous methods that relied on pruning,” not to the citing authors’ comparison of compression rates. The cited paper’s Contributions (p. 3) says its method uses pruning to uncover trainable subnetworks, so it fits that description. (LLM, conf=0.88)
- Supportedcites [15] Chen, 2022, Knowledge from the Original Network: Restore a Better Pruned Network with Knowledge Distillation, Complex Intell. Syst. source
The citation identifies a prior method combining pruning and knowledge distillation; the Introduction (p. 2) explicitly describes the paper’s method that way. The claim of greater compression is the citing authors’ comparison about their own approach, not a finding attributed to this source. (LLM, conf=0.92)
- Supportedcites [1] Hinton, G.E., Vinyals, O., and Dean, J. (2015). Distilling the Knowledge in a Neural Network. arXiv. source
The citation to Hinton et al. is attached to the phrase identifying previous methods that relied on knowledge distillation, not to the citing authors’ claim that their own approach achieves greater compression. The Introduction (pp. 1–2) describes distillation as transferring knowledge from a cumbersome model to a smaller one, so the paper is an example of the cited type of method. (LLM, conf=0.94)
- Not matchedcites [3] Tung, F., and Mori, G. (November, January 27). Similarity-Preserving Knowledge Distillation. Proceedings of the IEEE/CVF International Conference on Computer Vision, Seoul, Republic of Korea.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Supportedcites [5] Zagoruyko, S., and Komodakis, N. (2017). Paying More Attention to Attention: Improving the Performance of Convolutional Neural Networks via Attention Transfer. arXiv. source
The citation appears in the group identifying earlier knowledge-distillation methods, not as evidence for the citing authors’ own compression comparison. Section 3.1 says attention transfer can be combined with knowledge distillation, and Section 4.1.1 reports results for that combination. The source therefore supports its identification as an example of prior work using KD; it does not establish the citing authors’ claimed compression advantage. (LLM, conf=0.84)
- Supportedcites [7] Chen, D., Mei, J.-P., Zhang, Y., Wang, C., Feng, Y., and Chen, C. (2021, January 2–9). Cross-Layer Distillation with Semantic Calibration. Proceedings of the AAAI Conference on Artificial Intelligence, Online. source
The citation identifies a previous knowledge-distillation method; the paper’s §3.2 (p. 4) explicitly proposes “a new crosslayer knowledge distillation” method. The claim of greater compression belongs to the citing authors’ description of their own approach, not to this source. (LLM, conf=0.90)
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The results demonstrate the potential to compress the teacher model by over 70% on average, with minor adjustments in accuracy.”
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Notably, we demonstrate its efficacy by producing two compressed variants of ResNet 101: ResNet 50 and ResNet 18.”
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Notably, some cases even exhibit improved or consistent accuracy.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
23 more claims
Create a free account to see all 52 claims and every source verdict.
- Probably not a substantial claim. 3
- It may state two separate claims as one. 22
- Borderline: one of the tests for a substantial claim is unclear. 1
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
“| Compression rate | 8.9 × | 5.7 × | 13.9 ×”
Reported 5.7 · recomputed 16.27027
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 4.3. Distillation
“| SemCKD [7] SimKD [9] | 80.89 81.7 | 79.92 80.56 |”
§ 4.4. Pruning SimKD [9] 81.7
“| ACC(s) distillation | 81.7 | 80.1 | 80.65”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 2.1. Knowledge Distillation
“Self-Residual Representation Learning (SRRL), presented by Zhang et al. [6], proposes SRRL for knowledge distillation”
§ 7. Future Research
“6. Pham, T.X.; Niu, A.; Kang, Z.; Madjid, S.R.; Hong, J.W.; Kim, D.; Tee, J.T.J.; Yoo, C.D. Self-Supervised Visual Representation Learning via Residual Momentum.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 2.2. Pruning
“Chen et al. [15], introduce Rocket Launching, a general framework for training lightweight deep neural networks.”
§ 7. Future Research
“15. Chen, L.; Chen, Y.; Xi, J.; Le, X. Knowledge from the Original Network: Restore a Better Pruned Network with Knowledge Distillation.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 5. Discussion
“The crucial aspect of our study is pruning followed by KD.”
§ 6. Conclusions
“Our unique compression framework follows a sequential process: knowledge distillation, pruning of the distilled student model, and fine-tuning.”
Citations
Cites
Distilling the Knowledge in a Neural Network
Hinton, G., Vinyals, O., & Dean, J. (2015) · arXiv · doi:10.48550/arxiv.1503.02531 · Cited by 14,204
Han, S., Mao, H., & Dally, W. J. (2015) · arXiv · doi:10.48550/arxiv.1510.00149 · Cited by 4,662
Pruning Filters for Efficient ConvNets
Li, H., Kadav, A., Durdanovic, I., Samet, H., & Graf, H. P. (2016) · arXiv · doi:10.48550/arxiv.1608.08710 · Cited by 2,928
Zagoruyko, S., & Komodakis, N. (2016) · arXiv · doi:10.48550/arxiv.1612.03928 · Cited by 2,027
The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks
Frankle, J., & Carbin, M. (2018) · arXiv · doi:10.48550/arxiv.1803.03635 · Cited by 1,529 · cited 2×
“For instance, in the case of ResNet101, our compact version utilizes 0.32 × 10 7 parameters, whereas in the previous work [14], this count is 1.78 × 10 7 , and in another study [22], it is 1.73 × 10 7 .”
BackgroundMethodInfluentialContrastive Representation DistillationPreprint
Tian, Y., Krishnan, D., & Isola, P. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1910.10699 · Cited by 1,331
Similarity-Preserving Knowledge Distillation
Tung, F., & Mori, G. (2019) · 2019 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2019.00145 · Cited by 1,079
Importance Estimation for Neural Network Pruning
Molchanov, P., Mallya, A., Tyree, S., Frosio, I., & Kautz, J. (2019) · 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2019.01152 · Cited by 983 · cited 1×
“Molchanov et al. [11], introduce a method to compute the importance of each weight, considering both its magnitude and the layer’s sensitivity.”
MethodVariational Information Distillation for Knowledge Transfer
Ahn, S., Hu, S. X., Damianou, A., Lawrence, N. D., & Dai, Z. (2019) · 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2019.00938 · Cited by 671
Cross-Layer Distillation with Semantic Calibration
Chen, D., Mei, J.-P., Zhang, Y., Wang, C., Wang, Z., Feng, Y., & Chen, C. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i8.16865 · Cited by 335
Research works printed but not linked: 1
- LeCun, Y., Denker, J., and Solla, S. (1989). Proceedings of the Advances in Neural Information Processing Systems, Morgan-Kaufmann.
Cited by
Comparative analysis of model compression techniques for achieving carbon efficient AI
Paula, E., Soni, J., Upadhyay, H., & Lagos, L. E. (2025) · Scientific Reports · doi:10.1038/s41598-025-07821-w · Cited by 53
Matching the Ideal Pruning Method with Knowledge Distillation for Optimal Compression
Malihi, L., & Heidemann, G. (2024) · Applied System Innovation · doi:10.3390/asi7040056 · Cited by 18
Self-citationGenerative AI-Enabled Semantic Communication: State-of-the-Art, Applications, and the Way Ahead
Liang, C., & Li, D. (2025) · IEEE Communications Surveys & Tutorials · doi:10.1109/comst.2025.3649707 · Cited by 29
“Recent studies have demonstrated promising results in pruning strategies designed explicitly for semantic tasks [152].”
Federated Learning Optimization for Mobile Edge Devices using Knowledge Distillation and Pruning
Salaudeen, T. O., Razak, N. I. A., Abdullah, S. A. B. C., & Chairungsee, S. (2024) · IEEE Student Conference on Research and Development (SCOReD) · doi:10.1109/scored64708.2024.10872690 · Cited by 4
“[10] introduces an innovative framework that combines knowledge distillation, pruning and fine-tuning to achieve a compact model compression.”
InfluentialEnsemble learning for operations research and business analytics
Bock, K. W. D., Bogaert, M., & du Jardin, P. (2025) · Annals of Operations Research · doi:10.1007/s10479-025-06852-w · Cited by 5
“Other authors, on the other hand, focus on combining pruning and knowledge distillation in distributed systems (Malihi & Heidemann, 2023 ) to efficiently transfer knowledge from a pruned ensemble without com - promising the overall system performance.”
Knowledge Distillation Using Deep Learning Techniques: A Survey
Neeralgi, A. A., Avarsekar, R. R., Bhajantri, M., Khatod, R., Kulkarni, U., & Nadaf, S. M. (2024) · IEEE Conference on Engineering Informatics (ICEI) · doi:10.1109/icei64305.2024.10912425 · Cited by 2
“[7].”
Dynamic Relevance-Weighting-Based Width-Adaptive Auto-Encoder
Almejalli, M., Bchir, O., & Ismail, M. M. B. (2025) · Applied Sciences · doi:10.3390/app15126455 · Cited by 2
Evaluating Green AI Techniques in the Context of Cybercrime Prevention: A Review of 5G Networks
Shati, I. A. K., Gernaz, M. F. S., Jernz, R., & Arfa, A. S. A. (2026) · 2026 IEEE International conference on Advanced Systems and Emergent Technologies (IC_ASET) · doi:10.1109/ic_aset69920.2026.11502423 · Cited by 1
“Although promising, Green AI applications in 5G cybersecurity remain limited, underscoring the need for standardized metrics and deployment frameworks [12], [13].”
InfluentialWang, Z., Lin, S., & Zhou, C. H. (2025) · Egyptian Informatics Journal · doi:10.1016/j.eij.2025.100830 · Cited by 1
Jiang, W., Xu, G., & Song, A. (2025) · Applied Sciences · doi:10.3390/app15168839 · Cited by 1
Yuan, W., & Ge, Y. (2025) · GeoJournal · doi:10.1007/s10708-025-11325-5 · Cited by 1
“As a visual symbol bearing profound cultural connotation and artistic value, how to get innovative applications in modern design, especially in packaging design, has become a topic worthy of in-depth discussion (Malihi & Heidemann., 2023).”
Yagiz, M. A., Benamara, A., Deniz, Z., & Göktaş, P. (2026) · International Balkan Conference on Communications and Networking (BalkanCom) · doi:10.1109/balkancom71095.2026.11605017
“…resources, and data privacy requirements, by integrating three essential components: (1) a federated VAE tailored for anomaly detection in network traffic [19], (2) KD to create lightweight models suitable for edge deployment [9], and (3) privacy-preserving secure aggregation protocols [20].”
InfluentialJanardhanan, H. (2026) · doi:10.1109/iccces62661.2026.11436497
“Malihi & Heidemann [10] introduced a hybrid model compression framework that combined KD, network pruning and fine-tuning to enhance inference efficiency and controllable compactness.”
Ameh, J. E., Otebolaku, A., Shenfield, A., Ikpehai, A., & Sule, D. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-16089-8_12
Improving YOLOv11 Model for Fabric Defect Detection Using Knowledge Distillation Approach
The, S., Rawatan, J. E., Tanoto, A., et al. (2026) · Procedia Computer Science · doi:10.1016/j.procs.2026.08.299
Shi, J., Zhao, J., Qiang, Y., et al. (2025) · IEEE International Conference on Bioinformatics and Biomedicine (BIBM) · doi:10.1109/bibm66473.2025.11356652 · Cited by 1
“Compression techniques, such as knowledge distillation, pruning, and quantization, reduce model size by minimizing parameters or precision but often involve complex training and may degrade performance on resource-limited devices [14], [15].”
Farrukh, F. F., & Amayri, M. (2025) · Building and Environment · doi:10.1016/j.buildenv.2025.113836
Li, B. (2025) · Journal of Computational Methods in Sciences and Engineering · doi:10.1177/14727978251385188
Reducing Eye-Tracking Model Size for Web Applications with Pruning Methods
Nugroho, M. A., Abdurohman, M., Erfianto, B., et al. (2024) · doi:10.1109/icoict61617.2024.10698097
“Malihi and Heidemann (2023) found that pruned models exhibit superior accuracy compared to distilled models while utilizing fewer parameters [27].”
Metadata
- Authors
- Malihi, L., & Heidemann, G.
- Affiliations
- Osnabrück University (DE)
- Corresponding author
- Malihi, Leila
- ORCID
- Malihi, Leila
- Venue
- Big Data and Cognitive Computing
- Publisher
- MDPI AG
- Published
- 2023-09-19
- Volume
- 7
- Issue
- 3
- Pages
- 154
- ISSN
- 2504-2289 (ISSN-L)
- OpenAlex
- W4386895277
- Cited by
- 20
- References
- 14
- Citations by year
- 2026: 6 · 2025: 10 · 2024: 4
- FWCI
- 1.82
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Big Data and Cognitive Computing (publishedVersion, cc-by)
- osnaDocs (Osnabrück University) (submittedVersion, other-oa)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- osnaDocs (Osnabrück University) (cc-by)
- Funders
- Osnabrück University
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning and Data Classification
- Keywords
- model compression, knowledge distillation, network pruning, fine-tuning, ResNet, EfficientNet, DenseNet, CIFAR-10, CIFAR-100, parameter reduction, generalization
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-06, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-06 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.