Understanding deep learning (still) requires rethinking generalization
Open accessThe DOI matches its Crossref record; 45 of the 47 references it cites were found in a registry or an index; none of the 12 sources its claims cite is checked for support yet.
Abstract
Despite their massive size, successful deep artificial neural networks can exhibit a remarkably small gap between training and test performance. Conventional wisdom attributes small generalization error either to properties of the model family or to the regularization techniques used during training. Through extensive systematic experiments, we show how these traditional approaches fail to explain why large neural networks generalize well in practice. Specifically, our experiments establish that state-of-the-art convolutional networks for image classification trained with stochastic gradient methods easily fit a random labeling of the training data. This phenomenon is qualitatively unaffected by explicit regularization and occurs even if we replace the true images by completely unstructured random noise. We corroborate these experimental findings with a theoretical construction showing that simple depth two neural networks already have perfect finite sample expressivity as soon as the number of parameters exceeds the number of data points as it usually does in practice. We interpret our experimental findings by comparison with traditional models. We supplement this republication with a new section at the end summarizing recent progresses in the field since the original version of this paper.
Overview
CoChat Summary
TL;DR
Deep neural networks can fit random labels and random pixels with zero training error, so traditional complexity and regularization explanations do not account for their generalization. The paper argues that explicit regularization is neither necessary nor sufficient, and offers a finite-sample expressivity construction plus linear-model implicit-regularization analysis.
Key findings
- On CIFAR10, Inception, AlexNet, and MLPs reach 100% (or 99.82%/99.34%) training accuracy on random labels while test accuracy is near chance: Inception 9.78%, Inception w/o BatchNorm 10.12%, AlexNet 9.86%, MLP 3×512 10.48%, MLP 1×512 10.61%.
- On ImageNet, Inception reaches 95.20% top-1 accuracy on 1.2 million random labels from 1000 categories without hyperparameter tuning; with explicit regularizers on it still reaches ∼90% top-1.
- Explicit regularization may improve generalization but is neither necessary nor sufficient: on CIFAR10 all models still generalize well with regularizers off; on ImageNet top-1 accuracy without regularization is 59.80% vs 0.1% random guessing, and with data augmentation only Inception reaches 72.95% top-1.
- Theoretically, a two-layer ReLU network with p = 2n + d parameters can represent any labeling of any sample of size n in d dimensions; depth-k networks can use O(n/k) parameters per layer. In linear models, SGD converges to a minimum-norm solution, but minimum norm is not predictive: on MNIST the minimum norm with no preprocessing is ∼220 and with wavelet preprocessing ∼390, yet test error drops by a factor of 2.
Methods
- Randomization tests on CIFAR10 and ImageNet ILSVRC 2012: train Inception V3 on ImageNet and smaller Inception, AlexNet, and MLPs on CIFAR10 under true labels, partially corrupted labels, random labels, shuffled pixels, random pixels, and Gaussian pixels, toggling data augmentation, weight decay, and dropout.
- Compare training/test accuracy and convergence time; supplement with a finite-sample expressivity theorem for two-layer ReLU networks and with linear/kernel experiments on MNIST and CIFAR10 (Gaussian kernels, random convolutional filters).
Limitations
- Stated: the paper offers no new theory of generalization; it says a full mathematical characterization remains challenging and that it cannot fully survey recent progress. It also states that minimum-norm intuition is only a small piece of the generalization story and that it does not explain why certain architectures generalize better.
- Evident: the empirical scope is limited to image classification benchmarks (CIFAR10, ImageNet) and a small set of architectures/regularizers, so the findings may not extend to other domains or model families. The theoretical construction uses a depth-2 network with large width, and the paper notes depth-k alternatives with O(n/k) parameters per layer.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
0 Valid · 7 Not valid · 12 Not checked · 85 Own results and interpretations
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
“In other words, by randomizing labels alone we can force the generalization error of a model to jump up considerably without changing the model, its size, hyperparameters, or the optimizer.”
§ Understanding Deep Learning (Still) Requires Rethinking Generalization > 1.1. The randomization test
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.
None of the sources cited for this claim have been checked yet.
In the paper
“although it does not reach the perfect 100% top-1 accuracy, 95.20% accuracy is still very surprising for 1.2 million random labels from 1000 categories.”
§ Understanding Deep Learning (Still) Requires Rethinking Generalization > 2. EFFECTIVE CAPACITY OF NEURAL NETWORKS
plain- 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
“The networks fit the corrupted training set perfectly for all the cases.”
§ Understanding Deep Learning (Still) Requires Rethinking Generalization > Partially corrupted labels.
plain- 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
“On the CIFAR10 dataset, Alexnet and MLPs all converge to zero loss on the training set.”
§ Understanding Deep Learning (Still) Requires Rethinking Generalization > 2.1. Fitting random labels and pixels
plain- 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
“We corroborate these experimental findings with a theoretical construction showing that simple depth two neural networks already have perfect finite sample expressivity as soon as the number of parameters exceeds the number of data points as it usually does in practice.”
- Also stated in: § 1.3. Finite sample expressivity · plain
“More formally, we exhibit a very simple two-layer ReLU network with p = 2n + d parameters that can express any labeling of any sample of size n in d dimensions.”
§ Understanding Deep Learning (Still) Requires Rethinking Generalization > Abstract
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: § 1.3. Finite sample expressivity · plain
70 more claims
Create a free account to see all 104 claims and every source verdict.
- It may state two separate claims as one. 17
- Probably not a substantial claim. 15
- The quote only partly states this claim. 1
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Based on existing research(Saha et al., 2021; Zhang et al., 2021), the following conclusion can be drawn: the gradient updates of a linear layer always lie within the space spanned by its input.”
From Knowledge Decoupling via Orthogonal Projection for Lifelong Editing of Large Language Models (2025) · cites this paper as Zhang et al., 2021
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“If a model is successfully trained on a dataset with random labels, which has been shown to be possible with deep neural networks [294], then it has learned the data generation process by memorizing the random labels.”
From From Anecdotal Evidence to Quantitative Evaluation Methods: A Systematic Review on Evaluating Explainable AI (2023) · cites this paper as [294]
The check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“Existing works in privacy-preserving federated learning are mostly developed based on the wellknown privacy-preserving techniques, including: (1) homomorphic encryption (HE), such as Paillier [90], Elgamal [91] and Brakerski-Gentry-Vaikuntanathan cryptosystems [92]; (2) Secure Multiparty Computation (SMC), such as garbled circuits [93] and secret sharing [94]; and (3) differential privacy (DP) [95”
From Privacy and Robustness in Federated Learning: Attacks and Defenses (2024) · cites this paper as [95]
Experimental — accuracy not yet measured
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 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
§ Understanding Deep Learning (Still) Requires Rethinking Generalization > 2.1. Fitting random labels and pixels
“On the CIFAR10 dataset, Alexnet and MLPs all converge to zero loss on the training set.”
§ Understanding Deep Learning (Still) Requires Rethinking Generalization > Rademacher complexity and VC-dimension.
“(fitting random labels) No No 99.82 9.86 MLP 3 × 512 1,735,178 No Yes 100.0 53.35No No 100.0 52.39 (fitting random labels) No No 100.0 10.48 MLP 1 × 512 1,209,866 No Yes 99.80 50.39No No 100.0 50.51 (fitting random labels) No No 99.34 10.61”
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
§ Understanding Deep Learning (Still) Requires Rethinking Generalization > 1.1. The randomization test
“More precisely, when trained on a completely random labeling of the true data, neural networks achieve 0 training error.”
§ Understanding Deep Learning (Still) Requires Rethinking Generalization > 2. EFFECTIVE CAPACITY OF NEURAL NETWORKS
“although it does not reach the perfect 100% top-1 accuracy, 95.20% accuracy is still very surprising for 1.2 million random labels from 1000 categories.”
Citations
Cites
(2014) · Cited by 33,419
Wu, Y. (1999) · Technometrics · doi:10.1080/00401706.1999.10485951 · Cited by 27,221
Advances in neural information processing systems 7
Cios, K., & Shields, M. (1997) · Neurocomputing · doi:10.1016/s0925-2312(97)90018-7 · Cited by 22,059
Approximation by superpositions of a sigmoidal function
Cybenko, G. (1989) · Mathematics of Control, Signals, and Systems · doi:10.1007/bf02551274 · Cited by 13,883
Advances in Neural Information Processing Systems 25
Zhang, Y., Sutton, C. A., Storkey, A., & Zoubin Ghahramani, A. (2012) · Cited by 12,653
Understanding deep learning requires rethinking generalization
Zhang, C., Bengio, S., Hardt, M., Recht, B., & Vinyals, O. (2016) · arXiv · doi:10.48550/arxiv.1611.03530 · Cited by 2,838
Self-citationReconciling modern machine-learning practice and the classical bias–variance trade-off
Belkin, M., Hsu, D., Ma, S., & Mandal, S. (2019) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1903070116 · Cited by 1,595 · cited 1×
“23, 7, 6, 16 Despite significant progress on theoretical understanding of deep learning in the past few years, a full mathematical characterization of the whole story remains challenging.”
BackgroundA Generalized Representer Theorem
Schölkopf, B., Herbrich, R., & Smola, A. J. (2001) · Lecture Notes in Computer Science · doi:10.1007/3-540-44581-1_27 · Cited by 1,589 · cited 1×
“We have thus derived the “kernel trick” 34 —albeit in a round-about fashion.”
MethodA Closer Look at Memorization in Deep Networks
Arpit, D., Jastrzębski, S., Ballas, N., Krueger, D., Bengio, E., Kanwal, M. S., Maharaj, T., Fischer, A., Courville, A., Bengio, Y., et al. (2017) · arXiv · doi:10.48550/arxiv.1706.05394 · Cited by 1,307
Bartlett, P. L. (1998) · IEEE Transactions on Information Theory · doi:10.1109/18.661502 · Cited by 1,166
Research works printed but not linked: 8
- Cohen, N., Shashua, A. Convolutional rectifier networks as generalized tensor decompositions. In International Conference on Machine Learning. M.F. Balcan and K.Q. Weinberger, eds. 2016, 955--963.
- Cybenko, G. Approximation by superposition of sigmoidal functions. Math. Control Signal. Syst. 4, 2 (1989), 303--314.
- Golowich, N., Rakhlin, A., Shamir, O. Size-independent sample complexity of neural networks. In Conference On Learning Theory. P.R. Sébastien Bubeck, V. Perchet, eds., 2018, 297--299.
- Ma, S., Bassily, R., Belkin, M. The power of interpolation: Understanding the effectiveness of sgd in modern over-parametrized learning. In International Conference on Machine Learning. J. Dy and A. Krause, eds. 2018, 3325--3334.
- Morcos, A., Raghu, M., Bengio, S. Insights on representational similarity in neural networks with canonical correlation. Adv. Neural Inform. Process. Syst.. 2018, 5727--5736.
- Neyshabur, B., Tomioka, R., Srebro, N. Norm-based capacity control in neural networks. In Conference on Learning Theory. S.K. Peter Grünwald and E. Hazan, eds. 2015, 1376--1401.
- Smith, S.L., Le, Q.V. A bayesian perspective on generalization and stochastic gradient descent. In International Conference on Learning Representations, 2018.
- Zhou, W., Veitch, V., Austern, M., Adams, R.P., Orbanz, P. Non-vacuous generalization bounds at the ImageNet scale: A PAC-Bayesian compression approach. In International Conference on Learning Representations, 2019.
Cited by
Items 401–500 of 2,667
Document image analysis and recognition: a survey
Arlazarov, V. V., Control", C. S. A., Andreeva, E. I., Bulatov, K. B., Nikolaev, D., Petrova, O. O., Савельев, Б. И., & Slavin, O. (2022) · Computer Optics · doi:10.18287/2412-6179-co-1020 · Cited by 25
“Thus, a lot of modern architectures contain an excessive number of parameters and are prone to overfitting [102].”
BackgroundLiang, Y., Zhu, L., Wang, X., & Yang, Y. (2022) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2022.3213563 · Cited by 31
Hu, J., Yu, C., Zhang, Y., & Zhang, H. (2023) · International Journal of Neural Systems · doi:10.1142/s0129065723500600 · Cited by 26
Efficient IoT Devices Localization Through Wi-Fi CSI Feature Fusion and Anomaly Detection
Li, Y., Yang, J., Shih, S.-L., Shih, W.-T., Wen, C.-K., & Jin, S. (2024) · IEEE Internet of Things Journal · doi:10.1109/jiot.2024.3421577 · Cited by 26
“Considering deep neural networks’ propensity for overfitting on abnormal sequences, potentially impairing generalization [51], we enforce two criteria to ensure the training dataset predominantly contains normal sequences: • Criteria 1: Within each trajectory point, among the 5 paths identified by…”
MethodHigh-Dimensional Separability for One- and Few-Shot Learning
Gorban, A. N., Grechuk, B., Mirkes, E. M., Stasenko, S. V., & Tyukin, I. (2021) · Entropy · doi:10.3390/e23081090 · Cited by 31
“Two papers give a nice introduction in this area [19,20].”
BackgroundD'Orazio, R., Loizou, N., Laradji, I., & Mitliagkas, I. (2021) · Transactions on machine learning research · doi:10.48550/arxiv.2110.15412 · Cited by 36
“, they can fit or interpolate the training dataset (Zhang et al., 2021).”
BackgroundKanwal, N., Khoraminia, F., Kiraz, U., Mosquera‐Zamudio, A., Monteagudo, C. C., Janssen, E. A. M., Zuiverloon, T. C. M., Rong, C., & Engan, K. (2024) · BMC Medical Informatics and Decision Making · doi:10.1186/s12911-024-02676-z · Cited by 26
“It is a well-known problem that DL models suffer from poor generalization, robustness, and overconfident predictions over out-of-distribution (OoD) data [79– 81].”
BackgroundMethodTrustworthy learning with (un)sure annotation for lung nodule diagnosis with CT
Zhang, H., Chen, L., Gu, X., Zhang, M., Qin, Y., Yao, F., Wang, Z., Gu, Y., & Yang, G. (2022) · Medical Image Analysis · doi:10.1016/j.media.2022.102627 · Cited by 26
Early-Learning regularized Contrastive Learning for Cross-Modal Retrieval with Noisy Labels
Xu, T., Liu, X., Huang, Z. J., Guo, D., Hong, R., & Wang, M. (2022) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3503161.3548066 · Cited by 31
“In machine learning, handling noisy labeled data correctly is practically challenging as suggested in many previous works [20, 44].”
BackgroundRandom matrix analysis of deep neural network weight matrices
Thamm, M., Staats, M., & Rosenow, B. (2022) · Physical review. E · doi:10.1103/physreve.106.054124 · Cited by 52
“Network v) uses a technique called random cropping [34], i.”
BackgroundMethodInfluentialDynamic Mosaic algorithm for data augmentation
Li, Y., Cheng, R., Zhang, C., Chen, M., Liang, H., & Wang, Z. (2023) · Mathematical Biosciences & Engineering · doi:10.3934/mbe.2023311 · Cited by 25
Self-citationData augmentation for speech separation
Alex, A., Wang, L., Gastaldo, P., & Cavallaro, A. (2023) · Speech Communication · doi:10.1016/j.specom.2023.05.009 · Cited by 33
RankMatch: Fostering Confidence and Consistency in Learning with Noisy Labels
Zhang, Z., Chen, W., Fang, C., Li, Z., Chen, L., Lin, L., & Li, G. (2023) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51070.2023.00158 · Cited by 36
Enhanced machine learning models development for flash flood mapping using geospatial data
Hasnaoui, Y., Tachi, S. E., Bouguerra, H., Benmamar, S., Gilja, G., Szczepanek, R., Navarro-Pedreño, J., & Yaseen, Z. M. (2024) · Euro-Mediterranean Journal for Environmental Integration · doi:10.1007/s41207-024-00553-9 · Cited by 25
“Additionally, there is a notable difference between L1 and L2 regularization (Shanmugavel et al. 2023; Zhang et al. 2021).”
Yang, L., Liang, J., Liu, G., Jia, Y., Yang, S., Li, B., & Guo, Y. (2024) · Nano Energy · doi:10.1016/j.nanoen.2024.110591 · Cited by 26
ReSup: Reliable Label Noise Suppression for Facial Expression Recognition
Zhang, X., Lu, Y., Yan, H., Huang, J., Gu, Y., Ji, Y., Liu, Z., & Liu, B. (2025) · IEEE Transactions on Affective Computing · doi:10.1109/taffc.2025.3549017 · Cited by 35
“This is motivated by the memorization effect of deep neural networks (DNNs), where the model tends to memorize correctly labeled samples first [47, 3, 14], leading to noisy samples having higher loss during early epochs of training [2, 46].”
BackgroundInstructed Diffuser With Temporal Condition Guidance for Offline Reinforcement Learning
Hu, J., Sun, Y., Huang, S., Guo, S., Chen, H., Shen, L., Sun, L., Chang, Y., & Tao, D. (2026) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2026.3652193 · Cited by 29
“Thesecondchallenge istheslowlyevolvingdatadistribution,whichimpliesthatthe diffusionmodelisnolongertrainedonastaticdataset[77],[78].”
From Data Quality to Model Quality: An Exploratory Study on Deep Learning
He, T., Yu, S., Wang, Z., Li, J., & Chen, Z. (2019) · Proceedings of the 11th Asia-Pacific Symposium on Internetware · doi:10.1145/3361242.3361260 · Cited by 24
Deep Cross-Modal Hashing With Ranking Learning for Noisy Labels
Shu, Z., Bai, Y., Yong, K., & Yu, Z. (2024) · IEEE Transactions on Big Data · doi:10.1109/tbdata.2024.3423704 · Cited by 35
“Several studies [7], [8] have shown that hashing methods are easily misled by noisy labels, resulting in model performance degradation.”
Zhang, K., Yang, J., Huang, C., & Huai, X. L. (2024) · International Journal of Heat and Mass Transfer · doi:10.1016/j.ijheatmasstransfer.2024.126290 · Cited by 24
Suh, N., & Cheng, G. (2024) · Annual Review of Statistics and Its Application · doi:10.1146/annurev-statistics-040522-013920 · Cited by 47
Classical-quantum approach to image classification: Autoencoders and quantum SVMs
Slabbert, D., & Petruccione, F. (2025) · AVS Quantum Science · doi:10.1116/5.0261885 · Cited by 25
Spectral Bias in Practice: The Role of Function Frequency in Generalization
Fridovich-Keil, S., Gontijo-Lopes, R., & Roelofs, R. (2022) · neural information processing systems · doi:10.52202/068431-0535 · Cited by 60
Bai, J., Wang, Y., Xiao, Z. Q., & Alazab, M. (2022) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2022.3171349 · Cited by 24
“Zhang et al. [22] pointed out that a deep neural network can still fit the distribution of data and extract the information contained in the data when training the data containing noise.”
BackgroundEnhancing Federated Learning with In-Cloud Unlabeled Data
Wang, L., Xu, Y., Xu, H., Liu, J., Wang, Z., & Huang, L. (2022) · IEEE International Conference on Data Engineering (ICDE) · doi:10.1109/icde53745.2022.00015 · Cited by 33
Estimation of systolic blood pressure by Random Forest using heart sounds and a ballistocardiogram
González-Landaeta, R., Ramirez, B., & Mejia, J. (2022) · Scientific Reports · doi:10.1038/s41598-022-22205-0 · Cited by 29
Membership Inference Attacks Against Deep Learning Models via Logits Distribution
Yan, H., Li, S., Wang, Y., Zhang, Y., Sharif, K., Hu, H., & Li, Y. (2022) · IEEE Transactions on Dependable and Secure Computing · doi:10.1109/tdsc.2022.3222880 · Cited by 38
“(b)AR(Attack Recall), which is the training dataset member is correctly classified as member [27], [28], [29].”
BackgroundMethodTwo Coupled Rejection Metrics Can Tell Adversarial Examples Apart
Pang, T., Zhang, H., He, D., Dong, Y., Su, H., Chen, W., Zhu, J., & Liu, T. Y. (2022) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52688.2022.01479 · Cited by 41
AutoOLA: Automatic object level augmentation for wheat spikes counting
Zaji, A., Liu, Z., Xiao, G. G., Bhowmik, P., Sangha, J. S., & Ruan, Y. (2023) · Computers and Electronics in Agriculture · doi:10.1016/j.compag.2023.107623 · Cited by 24
Attentive-Adaptive Network for Hyperspectral Images Classification With Noisy Labels
Wang, L., Zhu, T., Kumar, N., Li, Z., Wu, C., & Zhang, P. (2023) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2023.3254159 · Cited by 24
“corrupted labels [23] than traditional classifiers.”
BackgroundDynamics-aware loss for learning with label noise
Li, X.-C., Xia, X., Zhu, F., Liu, T., Zhang, X.-Y., & Liu, C. (2023) · Pattern Recognition · doi:10.1016/j.patcog.2023.109835 · Cited by 29
Why Is Prompt Tuning for Vision-Language Models Robust to Noisy Labels?
Wu, C.-E. R., Tian, Y., Yu, H., Wang, H., Morgado, P., Hu, Y. H., & Yang, L. (2023) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51070.2023.01420 · Cited by 44
Noise-aware Learning from Web-crawled Image-Text Data for Image Captioning
Kang, W., Mun, J., Lee, S., & Roh, B. (2023) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51070.2023.00275 · Cited by 43
Enhancing Cardiovascular Disease Diagnosis: The Power of Optimized Ensemble Learning
Yazdi, F., & Asadi, S. (2025) · IEEE Access · doi:10.1109/access.2025.3550015 · Cited by 28
“Based on these optimal datasets, we train base classifiers, including Logistic Regression, KNeighbors Classifier, Decision Tree Classifier, Support Vector Classifier (SVC), and Gaussian Naive Bayes (Gaussian NB) [53], [54].”
Dirichlet-Based Prediction Calibration for Learning with Noisy Labels
Zong, C.-C., Wang, Y.-W., Xie, M.-K., & Huang, S.-J. (2024) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v38i15.29672 · Cited by 23
“Unlike traditional DNNs which give a point estimate of ρ , EDL regards ρ as a random variable and places a Dirichlet distribution over ρ to represent the probability density of each possible ρ .”
BackgroundInfluentialOraL: An Observational Learning Paradigm for Unsupervised Hyperspectral Change Detection
Wang, G., Zhang, X., Peng, Z., Tian, S., Zhang, T., Tang, X., & Jiao, L. (2025) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2025.3526960 · Cited by 28
“Inspired by the noisy learning theory [29], [30], [31], we propose an unsupervised hyperspectral change detection paradigm termed Observational Learning (OraL) , which consists of two sequential stages: observation and reproduction, as shown in Fig.”
InfluentialVirtual staining for pathology: Challenges, limitations and perspectives
Lin, W., Hu, Y., Zhu, R., Wang, B., & Wang, L. (2025) · Intelligent Oncology · doi:10.1016/j.intonc.2025.03.005 · Cited by 24
Deep-Learning-Based Detection of Segregations for Ultrasonic Testing
Elischberger, F., Bamberg, J., & Jiang, X. (2022) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2022.3144728 · Cited by 24
Malik, M. F., Burhan, Q.-U.-A., Akhtar, S., & Shafiq, M. (2022) · Business Process Management Journal · doi:10.1108/bpmj-04-2022-0168 · Cited by 28
CNLL: A Semi-supervised Approach For Continual Noisy Label Learning
Karim, N., Khalid, U., Esmaeili, A., & Rahnavard, N. (2022) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw56347.2022.00433 · Cited by 35
“[79] explain that deep neural networks can be efficiently trained with any ratio of noisy labels, but may generalize poorly on test samples.”
BackgroundSparse random neural networks for online anomaly detection on sensor nodes
Leroux, S., & Simoens, P. (2022) · Future Generation Computer Systems · doi:10.1016/j.future.2022.12.028 · Cited by 23
Niyogisubizo, J., Liao, L., Sun, Q., Nziyumva, E., Wang, Y., Luo, L., Lai, S., & Murwanashyaka, E. (2023) · International Journal of Intelligent Transportation Systems Research · doi:10.1007/s13177-023-00351-7 · Cited by 28
Do we really need a new theory to understand over-parameterization?
Oneto, L., Ridella, S., & Anguita, D. (2023) · Neurocomputing · doi:10.1016/j.neucom.2023.126227 · Cited by 40
Interpolating Graph Pair to Regularize Graph Classification
Guo, H., & Mao, Y. (2023) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v37i6.25941 · Cited by 33
“Consequently, regularization techniques have been actively proposed, aiming to empower the learning of GNNs while avoiding over-smoothing (Li, Han, and Wu 2018), oversquashing (Alon and Yahav 2021) and over-fitting (Zhang et al. 2021).”
BackgroundRegularly Truncated M-Estimators for Learning With Noisy Labels
Xia, X., Lu, P., Gong, C., Han, B., Yu, J., Yu, J., & Liu, T. (2023) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2023.3347850 · Cited by 30
TiDAL: Learning Training Dynamics for Active Learning
Kye, S. M., Choi, K., Byun, H., & Chang, B. (2023) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51070.2023.02041 · Cited by 32
“We further analyze that using our training dynamics prediction module is effective and the module successfully predicts the TD of unlabeled data.”
MethodPreserving Node-level Privacy in Graph Neural Networks
Xiang, Z., Wang, T., & Wang, D. (2024) · IEEE Symposium on Security and Privacy (S&P) · doi:10.1109/sp54263.2024.00270 · Cited by 43
“Modern machine learning/deep learning systems exhibit privacy risks: over-fitting by neural networks results in mem-orization of the training data [48] and leads to unacceptable privacy risk [36]; it is known that only sharing model updates still leads to privacy infringements as the adversary can…”
Learning Electromagnetic Metamaterial Physics With ChatGPT
Lu, D., Deng, Y., Malof, J. M., & Padilla, W. J. (2025) · IEEE Access · doi:10.1109/access.2025.3552418 · Cited by 24
“This so-called ‘‘data bottleneck’’ issue is compounded by questions of interpretability and understanding, and therefore explaining model predictions remains a significant hurdle [16], [17].”
The Curious Case of Convex Neural Networks
Sivaprasad, S., Singh, A., Manwani, N., & Gandhi, V. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-86486-6_45 · Cited by 41
Deep Metric Learning Based on Meta-Mining Strategy With Semiglobal Information
Jiang, X., Liu, S., Dai, X., Hu, G., Huang, X., Yao, Y., Xie, G.-S., & Shao, L. (2022) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2022.3202571 · Cited by 25
“However, it is known that neural networks learn the simple samples first and then the hard samples [22], [23].”
BackgroundTractability from overparametrization: the example of the negative perceptron
Montanari, A., Zhong, Y., & Zhou, K. (2024) · Probability Theory and Related Fields · doi:10.1007/s00440-023-01248-y · Cited by 37
Simple Lifelong Learning Machines
Vogelstein, J. T., Dey, J. K., Helm, H. S., LeVine, W., Mehta, R., Tomita, T. M., Xu, H., Geisa, A., Wang, Q., van de Ven, G. M., et al. (2025) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2025.3595364 · Cited by 29
“Capacity is defined as the size of the subset of hypotheses that is achievable by the learning algorithm [50].”
Intrinsic Dimension, Persistent Homology and Generalization in Neural NetworksPreprint
Birdal, T., Lou, A., Guibas, L. J., & Şimşekli, U. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2111.13171 · Cited by 99
A Close Look into the Calibration of Pre-trained Language Models
Chen, Y., Yuan, L., Cui, G., Liu, Z., & Ji, H. (2023) · Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2023.acl-long.75 · Cited by 22
Florensa, A. F., Armenteros, J. J. A., Nielsen, H., Aarestrup, F. M., & Clausen, P. T. L. C. (2024) · NAR Genomics and Bioinformatics · doi:10.1093/nargab/lqae106 · Cited by 32
Retinal Vessel Segmentation by a Transformer-U-Net Hybrid Model With Dual-Path Decoder
Zhang, Y., & Chung, A. C. S. (2024) · IEEE Journal of Biomedical and Health Informatics · doi:10.1109/jbhi.2024.3394151 · Cited by 26
“Under the assumption that the deep model learns easy and clean patterns before noisy patterns [42], we can use predictions from the model to correct annotation errors during the training process.”
MethodImproving Alzheimer’s disease classification using novel rewards in deep reinforcement learning
Hatami, M., Yaghmaee, F., & Ebrahimpour, R. (2024) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2024.106920 · Cited by 24
Few-shot learning for novel object detection in autonomous driving
Zhuang, Y., Liu, P., Yang, H., Zhang, K., Wang, Y., & Pu, Z. (2025) · Communications in Transportation Research · doi:10.1016/j.commtr.2025.100194 · Cited by 23
A novel intrinsic measure of data separability
Guan, S., & Loew, M. H. (2022) · Applied Intelligence · doi:10.1007/s10489-022-03395-6 · Cited by 23
“[4], the time to convergence for the same training loss on random labels on the CIFAR10 dataset was greater than when training on true labels.”
BackgroundOne-shot generalization in humans revealed through a drawing task
Tiedemann, H., Morgenstern, Y., Schmidt, F., & Fleming, R. W. (2022) · eLife · doi:10.7554/elife.75485 · Cited by 25
Quantum–Classical Image Processing for Scene Classification
Chalumuri, A., Kune, R., Kannan, S., & Manoj, B. S. (2022) · IEEE Sensors Letters · doi:10.1109/lsens.2022.3173253 · Cited by 22
Flooding-X: Improving BERT’s Resistance to Adversarial Attacks via Loss-Restricted Fine-Tuning
Liu, Q., Zheng, R., Rong, B., Liu, J., Liu, Z., Cheng, Z., Qiao, L., Gui, T., Zhang, Q., & Huang, X. (2022) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2022.acl-long.386 · Cited by 41
“The next stage 560 is when the training loss is decreasing but the 561 test loss is increasing, after which the training 562 loss continues to approach zero and memorize the 563 training data completely (Zhang et al., 2021; Belkin 564 et al., 2018; Arpit et al., 2017).”
BackgroundmedXGAN: Visual Explanations for Medical Classifiers through a Generative Latent Space
Dravid, A. V., Schiffers, F., Gong, B., & Katsaggelos, A. K. (2022) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw56347.2022.00331 · Cited by 28
“For instance, they perform poorly on data that is out-of-distribution with respect to their training set [40, 55].”
BackgroundNoise Is Also Useful: Negative Correlation-Steered Latent Contrastive Learning
Yan, J., Luo, L., Xu, C., Deng, C. D., & Huang, H. (2022) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52688.2022.00013 · Cited by 27
“Although DNNs have high model capacities, they often overfit the noisy labels due to the memorization effect, resulting in poor classification and generalization performance [53].”
BackgroundDeep Blind Image Quality Assessment Powered by Online Hard Example Mining
Wang, Z., Jiang, Q., Zhao, S., Feng, W., & Lin, W. (2023) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2023.3257564 · Cited by 27
“Our initial rationale arises from [15], [16] that DNNs tend to first memorize easy patterns shared by the majority of training samples and then fit the hard and underrepresented patterns.”
BackgroundStrategizing AI in Business and Education: Emerging Technologies and Business Strategy
Przegalińska, A., & Jemielniak, D. (2023) · Cambridge University Press eBooks · doi:10.1017/9781009243520 · Cited by 28
Infrared-Visible Synthetic Data from Game Engine for Image Fusion Improvement
Gu, X., Liu, G., Zhang, X., Tang, L., Zhou, X., & Qiu, W. (2023) · IEEE Transactions on Games · doi:10.1109/tg.2023.3263001 · Cited by 22
“However, most of these traditional fusion methods rely on the artificial de-sign of fusion rules, which is not only computationally complex but also has some limitations in generalization [9], [10].”
BackgroundFrom Grim Reality to Practical Solution: Malware Classification in Real-World Noise
Wu, X., Guo, W., Yan, J., Coşkun, B., & Xing, X. (2023) · IEEE Symposium on Security and Privacy (S&P) · doi:10.1109/sp46215.2023.10179453 · Cited by 34
An overview of ensemble and feature learning in few-shot image classification using siamese networks
Valero-Mas, J. J., Gallego, A., & Rico-Juan, J. R. (2023) · Multimedia Tools and Applications · doi:10.1007/s11042-023-15607-3 · Cited by 28
“Nevertheless, the main counterpart of the supervised DL framework is the high amount of data required for training the model [30, 55].”
How to Prevent the Poor Performance Clients for Personalized Federated Learning?
Qu, Z., Li, X., HAN, X. J., Duan, R., Shen, C., & Chen, L. (2023) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52729.2023.01171 · Cited by 22
“Recently, some studies in centralized learning [17,51] observe that generalization is positively correlated to the sharpness of the training loss landscape.”
BackgroundLearning From Noisy Correspondence With Tri-Partition for Cross-Modal Matching
Feng, Z., Zeng, Z., Guo, C., Li, Z., & Hu, L. (2023) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2023.3318002 · Cited by 33
Transformer-Based Annotation Bias-Aware Medical Image Segmentation
Liao, Z., Hu, S., Xie, Y., & Xia, Y. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-43901-8_3 · Cited by 22
Tran, V. N., Ivanov, V. Y., & Kim, J. (2023) · Advances in Water Resources · doi:10.1016/j.advwatres.2023.104569 · Cited by 21
Understanding Deep Learning via Decision Boundary
Lei, S., He, F., Yuan, Y., & Tao, D. (2023) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2023.3326654 · Cited by 30
“Zhang et al. [12] demonstrate that neural networks can near-perfectly fit noisy labels (which suggests that deep learning has an extremely large Rademarcher complexity), but still have impressive generalization performance.”
BackgroundBayesian DivideMix++ for Enhanced Learning with Noisy Labels
Nagarajan, B., Marques, R., Aguilar, E., & Radeva, P. I. (2024) · Neural Networks · doi:10.1016/j.neunet.2024.106122 · Cited by 23
CTRL: Clustering Training Losses for Label Error Detection
Yue, C., & Jha, N. K. (2024) · IEEE Transactions on Artificial Intelligence · doi:10.1109/tai.2024.3365093 · Cited by 28
“This phenomenon is independent of the optimizations used during training or the NN architectures employed (Zhang et al. 2021).”
BackgroundMethodLiu, C., Albrecht, C. M., Wang, Y., Li, Q., & Zhu, X. X. (2024) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2024.3373908 · Cited by 33
“As reported in [8], deep learning models are known for their large number of parameters and capability of learning complex functions, yet vulnerability to label noise.”
BackgroundAdaptive Plasticity Improvement for Continual Learning
Liang, Y.-S., & Li, W.-J. (2023) · 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52729.2023.00755 · Cited by 21
“According to the existing works [24, 31, 36], the following proposition holds:”
BackgroundOptimizing quantum convolutional neural network architectures for arbitrary data dimension
Lee, C. W., Araujo, I. F., Kim, D.-H., Lee, J.-H., Park, S., Ryu, J.-Y., & Park, D. K. (2025) · Frontiers in Physics · doi:10.3389/fphy.2025.1529188 · Cited by 24
“Although the mathematical foundation for the success of DNNs remains an active area of research [36], several studies, as well as the universal approximation theorem, have demonstrated that neural networks can approximate complex functions with arbitrary accuracy [37].”
Incubating advances in integrated photonics with emerging sensing and computational capabilities
Jain, S., Hlaing, M. H., Fan, K.-C., Midkiff, J., Ning, S., Feng, C., Hsiao, P.-Y., Camp, P. T., & Chen, R. T. (2025) · Applied Physics Reviews · doi:10.1063/5.0218564 · Cited by 21
Evaluating the Robustness of Trigger Set-Based Watermarks Embedded in Deep Neural Networks
Lee, S., Song, W., Jana, S., Cha, M., & Son, S. (2022) · IEEE Transactions on Dependable and Secure Computing · doi:10.1109/tdsc.2022.3196790 · Cited by 27
Assessing the Experience-Sharing Parenting Method through Online Learning during Covid-19 Pandemic
Universitas Negeri Yogyakarta, I., Hardianto, D., Budiningsih, C. A., Universitas Negeri Yogyakarta, I., Haryanto, H., Universitas Negeri Yogyakarta, I., Pratama, K. W., Engineering, A. E. (., Ali, S. K. S., & Karakauki, M. (2022) · International Journal of Instruction · doi:10.29333/iji.2022.15420a · Cited by 21
Deep empirical risk minimization in finance: Looking into the future
Reppen, M., & Soner, H. M. (2022) · Mathematical Finance · doi:10.1111/mafi.12365 · Cited by 32
“Hence, as opposed to interesting recent studies [11, 55] arguing the benefits of more complex networks and interpolation, in this dynamic setting, overfitting causes the loss of the most salient restriction of the problem, namely, adaptedness of decisions to the information flow.”
BackgroundContext Label Learning: Improving Background Class Representations in Semantic Segmentation
Li, Z., Kamnitsas, K., Ouyang, C., Chen, C., & Glocker, B. (2023) · IEEE Transactions on Medical Imaging · doi:10.1109/tmi.2023.3242838 · Cited by 24
“Their large number of trainable parameters make them capable to fit to different kinds of tasks yielding high performance in terms segmentation accuracy [37].”
BackgroundRank-Aware Negative Training for Semi-Supervised Text Classification
Ahmed, M., Pan, S., Bo, W., Su, J., Cao, X., Zhang, W., & Liu, Y. (2023) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00574 · Cited by 24
“However, the pseudo-labels are treated equally likely to the truth labels and thus may lead to the error accumulation (Zhang et al., 2021; Arazo et al., 2020).”
BackgroundPrestressing wire breakage monitoring using sound event detection
Farhadi, S., Corrado, M., Borla, O., & Ventura, G. (2023) · Computer-Aided Civil and Infrastructure Engineering · doi:10.1111/mice.13079 · Cited by 24
“…such as weights and bias, to train an ANN model is exceptionally high; therefore, to build a generalized model, a large number of training data sets is required for each class to cover acoustic class variability based on the model complexity (Takahashi et al., 2016; C. Zhang et al., 2021).”
BackgroundRDGSL: Dynamic Graph Representation Learning with Structure Learning
Zhang, S., Xiong, Y., Zhang, Y., Sun, Y., Chen, X., Jiao, Y., & Zhu, Y. (2023) · ACM International Conference on Information and Knowledge Management (CIKM) · doi:10.1145/3583780.3615023 · Cited by 31
“We conduct a preliminary experiment on the Reddit dataset [12] to test how noise impacts the link prediction results of TGNs in Figure 1.”
MethodInfluentialNasir, T., & Malik, M. K. (2023) · Information Processing & Management · doi:10.1016/j.ipm.2023.103544 · Cited by 21
Yu, C., Zhang, D., Zou, W. H., & Li, M. (2024) · IEEE Transactions on Affective Computing · doi:10.1109/taffc.2024.3382618 · Cited by 22
“Unreliable labels may lead to overfitting and poor generalization [29].”
BackgroundVisual–language foundation models in medicine
Liu, C., Jin, Y., Guan, Z., Li, T., Qin, Y., Qian, B., Jiang, Z., Wu, Y.-L., Wang, X., Zheng, Y. F., et al. (2024) · The Visual Computer · doi:10.1007/s00371-024-03579-w · Cited by 44
“The vulnerability of expressive AI architectures, notably deep neural networks, to overfit-ting poses a significant hurdle [120, 121].”
Robust Noisy Correspondence Learning with Equivariant Similarity Consistency
Yang, Y., Wang, L., Yang, E., & Deng, C. D. (2024) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52733.2024.01676 · Cited by 39
“Relevant methods like [15, 29, 49] are almost based on the memorization effect of DNNs [51] observed in [13], which means that DNNs often prioritize learning simple patterns over fitting noisy samples.”
KAYA, Y. Ç. (2024) · Diagnostics · doi:10.3390/diagnostics14192253 · Cited by 21
Raghu, S. T. P., MacIsaac, D., & Scheme, E. (2024) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2024.109479 · Cited by 23
“This lack of unequivocal labeling can lead to detrimental consequences, as deep learning models are known for their ability to memorize even random associations [25], potentially hindering their ability to learn meaningful representations and generalize to real-world scenarios.”
BackgroundReLU Neural Networks with Linear Layers Are Biased towards Single- and Multi-index Models
Parkinson, S., Ongie, G., & Willett, R. M. (2025) · SIAM Journal on Mathematics of Data Science · doi:10.1137/24m1672158 · Cited by 28
“…networks, it has been argued that “the size [magnitude] of the weights is more important than the size [number of weights or parameters] of the network" [1], an idea reinforced by [2, 3] and yielding insight into the generalization performance of overparameterized neural networks [4, 5, 6, 7].”
Training Fully Connected Neural Networks is $\exists\mathbb{R}$-Complete
Bertschinger, D., Hertrich, C., Jungeblut, P., Miltzow, T., & Weber, S. (2026) · Theory of Computing · doi:10.4086/toc.2026.v022a006 · Cited by 24
Nunavath, V., Goodwin, M., Fidje, J. T., & Moe, C. E. (2018) · Communications in computer and information science · doi:10.1007/978-3-319-98204-5_18 · Cited by 26
Gao, J., Zuo, F., Özbay, K. M. A., Hammami, O., & Barlas, M. L. (2022) · Transportation Research Part A Policy and Practice · doi:10.1016/j.tra.2022.05.024 · Cited by 31
Information-Theoretic Characterizations of Generalization Error for the Gibbs Algorithm
Aminian, G., Bu, Y., Toni, L., Rodrigues, M. R. D., & Wornell, G. W. (2023) · IEEE Transactions on Information Theory · doi:10.1109/tit.2023.3329617 · Cited by 30
“As a consequence, both methods fail to capture the fact that the generalization error depends strongly on the interplay between the hypothesis class, learning algorithm, and the underlying data-generating distribution, as discussed in [7], [9].”
BackgroundTopological Data Analysis for Neural Network Analysis: A Comprehensive SurveyPreprint
Ballester, R., Casacuberta, C., & Escalera, S. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2312.05840 · Cited by 29
Curriculum-Based Federated Learning for Machine Fault Diagnosis With Noisy Labels
Sun, W., Yan, R., Jin, R., Zhao, R., & Chen, Z. (2024) · IEEE Transactions on Industrial Informatics · doi:10.1109/tii.2024.3435449 · Cited by 20
Metadata
- Authors
- Zhang, C., Bengio, S., Hardt, M., Recht, B., & Vinyals, O.
- Affiliations
- Google (United States) (US)
- Google Brain, Mountain View, CA
- University of California, Berkeley (US)
- Google DeepMind (United Kingdom) (GB)
- Venue
- Communications of the ACM
- Published
- 2021-02-22
- Volume
- 64
- Issue
- 3
- Pages
- 107-115
- ISSN
- 0001-0782 (ISSN-L)
- 1557-7317
- OpenAlex
- W2566079294
- Cited by
- 2,400
- References
- 47
- Citations by year
- 2026: 295 · 2025: 558 · 2024: 579 · 2023: 500 · 2022: 316 · 2021: 108
- FWCI
- 132.37
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- Communications of the ACM (publishedVersion, cc-by-nc-sa)
- Funders
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Stochastic Gradient Optimization Techniques
- Physical Sciences › Computer Science › Artificial Intelligence › Neural Networks and Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Keywords
- generalization, deep neural network, overfitting, overparameterization, expressivity, stochastic gradient descent, regularization, image classification, convolutional neural networks
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.
A language model drew the claims from the paper on 2026-11-08, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-11-08 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
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.