Deep Learning Under Privileged Information Using Heteroscedastic Dropout
Open accessAbstract
Unlike machines, humans learn through rapid, Abstract model-building. The role of a teacher is not simply to hammer home right or wrong answers, but rather to provide intuitive comments, comparisons, and explanations to a pupil. This is what the Learning Under Privileged Information (LUPI) paradigm endeavors to model by utilizing extra knowledge only available during training. We propose a new LUPI algorithm specifically designed for Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs). We propose to use a heteroscedastic dropout (i.e. dropout with a varying variance) and make the variance of the dropout a function of privileged information. Intuitively, this corresponds to using the privileged information to control the uncertainty of the model output. We perform experiments using CNNs and RNNs for the tasks of image classification and machine translation. Our method significantly increases the sample efficiency during learning, resulting in higher accuracy with a large margin when the number of training examples is limited. We also theoretically justify the gains in sample efficiency by providing a generalization error bound decreasing with O(1/n), where n is the number of training examples, in an oracle case.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
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?
In 2019, Xue et al. [640] used ResNet-101 with Ensemble for classification of skin cancer on 1 dataset.
“Xue al. [640] | et | 2019 | ResNet-101 | Classification | Skin cancer | 1 | Ensemble”
From A review of uncertainty quantification in deep learning: Techniques, applications and challenges (2021) · cites this paper as [640]
The Datasets section says the paper's experiments use ImageNet and Multi-30K, not a skin-cancer dataset. Its Heteroscedastic Dropout Implementation section describes VGG-16 and ResNet-50, not a ResNet-101 ensemble. The cited paper therefore describes a different study from the Xue et al. entry. (LLM, conf=0.86)
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 1× · see citing works
“[15] and a decay of 1 × 10−4 was used for image classification.”
MethodAdam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946 · cited 5× · see citing works
“We use the ADAM [21] optimizer in PyTorch for both image clas-sification and multi-modal machine translation.”
InfluentialImageNet: A large-scale hierarchical image database
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009) · 2009 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2009.5206848 · Cited by 65,333 · cited 8× · see citing works
“In this section, we provide two experimental results missing in the paper: first, an analysis of accuracy vs. the amount of x (cid:63) provided for image classification and second, a comparison of our method with baselines for the task of ImageNet image classification with 200 K images.”
Influential(2014) · Cited by 33,419 · see citing works
Auto-Encoding Variational Bayes
Kingma, D. P., & Welling, M. (2013) · arXiv · doi:10.48550/arxiv.1312.6114 · Cited by 20,067 · see citing works
Neural Machine Translation by Jointly Learning to Align and Translate
Bahdanau, D., Cho, K., & Bengio, Y. (2014) · arXiv · doi:10.48550/arxiv.1409.0473 · Cited by 16,213 · cited 1× · see citing works
“We use attention [2] and feed the image as a feature vector computed using the VGG[32] architecture pre-trained on ImageNet.”
Dropout as a bayesian approximation: Representing model uncertainty in deep learning
Gal, Y., & Ghahramani, Z. (2015) · arXiv · doi:10.48550/arxiv.1506.02142 · Cited by 6,471 · cited 1× · see citing works
“The rationale behind heteroscedastic dropout follows the close relationship between Bayesian learning and dropout presented by Gal and Gharamani[13].”
Moses: open source toolkit for statistical machine translation
Koehn, P., Zens, R., Dyer, C., Bojar, O., Constantin, A., Herbst, E., Hoang, H., Birch, A., Callison-Burch, C., Federico, M., et al. (2007) · Proceedings of the 45th Annual Meeting of the ACL on Interactive Poster and Demonstration Sessions - ACL '07 · doi:10.3115/1557769.1557821 · Cited by 4,735 · see citing works
Learning Deconvolution Network for Semantic Segmentation
Noh, H., Hong, S., & Han, B. (2015) · 2015 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2015.178 · Cited by 4,422 · cited 3× · see citing works
“We use attention [2] and feed the image as a feature vector computed us-ing the VGG[34] architecture pre-trained on ImageNet.”
Bleu: a method for automatic evaluation of machine translation
Papineni, K. (2024) · doi:10.57702/ljteqmrb · Cited by 3,682 · see citing works
Deep learning and the information bottleneck principle
Tishby, N., & Zaslavsky, N. (2015) · Information Theory Workshop · doi:10.1109/itw.2015.7133169 · Cited by 2,170 · cited 2× · see citing works
“Although we use IB [35], Gaussian dropout [33] and the re-parametrization trick [21], we are the first to our knowledge to apply any of these methods to the LUPI problem.”
Meteor Universal: Language Specific Translation Evaluation for Any Target Language
Denkowski, M., & Lavie, A. (2014) · WMT@ACL · doi:10.3115/v1/w14-3348 · Cited by 2,108 · cited 3× · see citing works
“Multi-modal Machine Translation Our method results in a significant accuracy improvement measure by both BLEU and METEOR scores.”
OpenNMT: Open-Source Toolkit for Neural Machine Translation
Klein, G., Kim, Y., Deng, Y., Sénellart, J., & Rush, A. M. (2017) · Annual Meeting of the Association for Computational Linguistics: System Demonstrations (ACL Demo) · doi:10.18653/v1/p17-4012 · Cited by 1,771 · cited 3× · see citing works
“Our LSTM implementation directly follows the baseline implementation provided by OpenNMT [23].”
MethodInfluentialA theoretically grounded application of dropout in recurrent neural networks
Gal, Y., & Zoubin Ghahramani, A. (2016) · Cambridge University Engineering Department Publications Database · Cited by 1,051 · see citing works
A new learning paradigm: Learning using privileged information
Vapnik, V., & Vashist, A. (2009) · Neural Networks · doi:10.1016/j.neunet.2009.06.042 · Cited by 774 · cited 10× · see citing works
“Learning Under Privileged Information: Learning under Privileged Information (LUPI) is initially proposed by Vapnik and Vashist [38, 37].”
InfluentialVariational Dropout and the Local Reparameterization TrickPreprint
Kingma, D. P., Salimans, T., & Welling, M. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1506.02557 · Cited by 708 · cited 1× · see citing works
“Its variational extension [22] uses local reparameterization to perform Bayesian learning.”
MethodXu, H., & Mannor, S. (2011) · Machine Learning · doi:10.1007/s10994-011-5268-1 · Cited by 389 · cited 2× · see citing works
“Moreover, any learning algorithm with a Lipschitz loss function admits the following result [39]; Proposition 1 ([39, Example 4]).”
BackgroundInformation dropout: learning optimal representations through noisy computation
Achille, A., & Soatto, S. (2018) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2017.2784440 · Cited by 356 · cited 18× · see citing works
“Heteroscedastic Dropout Implementation: We set λ = 100 in all experiments, although we found this was not a meaningful hyperparameter.”
InfluentialUnifying distillation and privileged information
Lopez-Paz, D., Bottou, L., Schölkopf, B., & Vapnik, V. (2015) · arXiv · doi:10.48550/arxiv.1511.03643 · Cited by 270 · cited 1× · see citing works
“In contrast to max-margin methods, the literature on learning CNNs under privileged information heavily uses the distillation framework, following the close relationship between distillation and LUPI studied in [25].”
Learning with Side Information through Modality Hallucination
Hoffman, J., Gupta, S., & Darrell, T. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.96 · Cited by 207 · cited 4× · see citing works
“Modality Hallucination[18]: Distillation-based LUPI method designed for multi-modal CNNs.”
InfluentialLearning to Rank Using Privileged Information
Sharmanska, V., Quadrianto, N., & Lampert, C. H. (2013) · IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2013.107 · Cited by 196 · cited 1× · see citing works
“This method is further applied to various computer vision problems [31, 29, 11] as well as ranking [30], clustering [10] and metric learning [12] problems.”
Doubly-Attentive Decoder for Multi-modal Neural Machine Translation
Calixto, I., Liu, Q., & Campbell, N. (2017) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/p17-1175 · Cited by 185 · see citing works
Incorporating Global Visual Features into Attention-based Neural Machine Translation.
Calixto, I., & Liu, Q. (2017) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/d17-1105 · Cited by 160 · cited 1× · see citing works
“A LUPI model [9] was already shown to perform better than multi-modal translation models which can use both images and sentences in test time [3, 4, 17, 36].”
MethodGeneralization with Deep Learning
Chen, Z., Wu, M., & Li, X. (2020) · WORLD SCIENTIFIC eBooks · doi:10.1142/11784 · Cited by 152 · see citing works
Representation of Linguistic Form and Function in Recurrent Neural Networks
Kádár, Á., Chrupała, G., & Alishahi, A. (2017) · Computational Linguistics · doi:10.1162/coli_a_00300 · Cited by 142 · cited 1× · see citing works
“Analysis of this model [5, 19] suggests that the embeddings learned by using vision as a privileged information are significantly different than language only ones and correlate better with human judgments.”
Multimodal Pivots for Image Caption Translation
Hitschler, J., Schamoni, S., & Riezler, S. (2016) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/p16-1227 · Cited by 94 · cited 1× · see citing works
“A LUPI model [9] was already shown to perform better than multi-modal translation models which can use both images and sentences in test time [3, 4, 17, 36].”
Incorporating Privileged Information Through Metric Learning
Fouad, S., Peter Tiňo, B., Raychaudhury, S., & Schneider, P. (2013) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2013.2251470 · Cited by 78 · cited 1× · see citing works
“This method is further applied to various computer vision problems [31, 29, 11] as well as ranking [30], clustering [10] and metric learning [12] problems.”
Privileged information for data clustering
Feyereisl, J., & Aickelin, U. (2011) · Information Sciences · doi:10.1016/j.ins.2011.04.025 · Cited by 75 · cited 1× · see citing works
“This method is further applied to various computer vision problems [31, 29, 11] as well as ranking [30], clustering [10] and metric learning [12] problems.”
Yang, H., Zhou, J. T., Cai, J., & Ong, Y.-S. (2017) · 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/cvpr.2017.635 · Cited by 75 · cited 3× · see citing works
“MIML-FCN[44]: a CNN-based LUPI framework designed for multi-instance problems.”
InfluentialInformation Bottleneck Learning Using Privileged Information for Visual Recognition
Motiian, S., Piccirilli, M., Adjeroh, D., & Doretto, G. (2016) · 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/cvpr.2016.166 · Cited by 59 · cited 6× · see citing works
“For the case of SVMs, Vapnik et al. [42] showed that utilizing privileged information can result in a generalization error bound with rate where n is the dataset size.”
InfluentialLearning language through pictures
Chrupała, G., Kádár, Á., & Alishahi, A. (2015) · doi:10.3115/v1/p15-2019 · Cited by 50 · see citing works
Training Group Orthogonal Neural Networks with Privileged Information
Chen, Y., Jin, X., Feng, J., et al. (2017) · International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2017/212 · Cited by 41 · cited 4× · see citing works
“Group Orthogonal neural networks [45] follow this approach.”
InfluentialObject Localization based on Structural SVM using Privileged Information
Feyereisl, J., Kwak, S., Son, J., & Han, B. (2014) · Cited by 39 · see citing works
Incorporating Global Visual Features into Attention-Based Neural Machine TranslationPreprint
Calixto, I., Liu, Q., & Campbell, N. J. H. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1701.06521 · Cited by 28 · cited 1× · see citing works
“A LUPI model [9] was already shown to perform better than multi-modal translation models which can use both images and sentences in test time [3, 4, 17, 36].”
MethodMind the Nuisance: Gaussian Process Classification using Privileged NoisePreprint
Hernández-Lobato, D., Sharmanska, V., Kristian Kersting, B., et al. (2014) · arXiv (Cornell University) · doi:10.48550/arxiv.1407.0179 · Cited by 20 · cited 2× · see citing works
“One closely related work is [16], extending Gaussian processes to the LUPI paradigm.”
Learning to Transfer Privileged InformationPreprint
Sharmanska, V., Quadrianto, N., & Lampert, C. H. (2014) · arXiv (Cornell University) · doi:10.48550/arxiv.1410.0389 · Cited by 18 · cited 1× · see citing works
“This method is further applied to various computer vision problems [31, 29, 11] as well as ranking [30], clustering [10] and metric learning [12] problems.”
Neural Machine Translation with Latent Semantic of Image and TextPreprint
Toyama, J., Misono, M., Suzuki, M., et al. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1611.08459 · Cited by 15 · cited 1× · see citing works
“A LUPI model [9] was already shown to perform better than multi-modal translation models which can use both images and sentences in test time [3, 4, 17, 36].”
Information Dropout: Learning Optimal Representations Through Noisy ComputationPreprint
Achille, A., & Soatto, S. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1611.01353 · Cited by 13 · cited 18× · see citing works
“Heteroscedastic Dropout Implementation: We set λ = 100 in all experiments, although we found this was not a meaningful hyperparameter.”
InfluentialLearning language through picturesPreprint
Chrupała, G., Kádár, Á., & Alishahi, A. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1506.03694 · Cited by 8 · see citing works
Doubly-Attentive Decoder for Multi-modal Neural Machine TranslationPreprint
Calixto, I., Liu, Q., & Campbell, N. J. H. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1702.01287 · Cited by 6 · see citing works
In the Era of Deep Convolutional Features: Are Attributes Still Useful Privileged Data?
Sharmanska, V., & Quadrianto, N. (2017) · Advances in computer vision and pattern recognition · doi:10.1007/978-3-319-50077-5_3 · Cited by 3 · cited 1× · see citing works
“This method is further applied to various computer vision problems [31, 29, 11] as well as ranking [30], clustering [10] and metric learning [12] problems.”
Method
Flagged first, then by citations
Research works printed but not linked: 5
- simonyan, 2014, Very Deep Convolutional Networks for Large-scale Image Recognition
- vapnik, 2015, Learning using privileged information: Similarity control and knowledge transfer, Journal of Machine Learning Research, 16, 2023
- tishby, 2000, The information bottleneck method
- elliott, 2016, Multi30k Multilingual english-german image descriptions, 70
- elliott, 2017, Imagination improves multimodal translation
Cited by
Sorted by citations
A review of uncertainty quantification in deep learning: Techniques, applications and challenges
Abdar, M., Pourpanah, F., Hussain, S., Rezazadegan, D., Liu, L., Ghavamzadeh, M., Fieguth, P., Cao, X., Khosravi, A., Acharya, U. R., et al. (2021) · Information Fusion · doi:10.1016/j.inffus.2021.05.008 · Cited by 2,720 · see citing works
A Survey on Information Bottleneck
Hu, S., Lou, Z., Yan, X., & Ye, Y. (2024) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2024.3366349 · Cited by 114 · see citing works
SimAug: Learning Robust Representations from Simulation for Trajectory Prediction
Liang, J., Jiang, L., & Hauptmann, A. G. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58601-0_17 · Cited by 100 · see citing works
“Methods for learning using privileged information [30,66,39,41] is not applicable for a similar reason.”
BackgroundMultimodal Machine Learning in Image-Based and Clinical Biomedicine: Survey and Prospects
Warner, E., Lee, J., Hsu, W. W. Y., Syeda-Mahmood, T. F., Kahn, C. E., Gevaert, O., & Rao, A. U. K. (2024) · International Journal of Computer Vision · doi:10.1007/s11263-024-02032-8 · Cited by 93 · see citing works
“Healthcare professionals and physicians already embrace the concept of multi-modal cognitive models in their diagnostic and prognostic practices, signaling that such computer models based on multimodal frameworks are likely to endure within the biomedical landscape.”
BackgroundPiotrowski, A., Napiorkowski, J. J., & Piotrowska, A. E. (2019) · Earth-Science Reviews · doi:10.1016/j.earscirev.2019.103076 · Cited by 81 · see citing works
Learning the implicit strain reconstruction in ultrasound elastography using privileged information
Gao, Z., Wu, S., Liu, Z., Luo, J., Zhang, H., Gong, M., & Li, S. (2019) · Medical Image Analysis · doi:10.1016/j.media.2019.101534 · Cited by 73 · see citing works
Learning Joint Visual Semantic Matching Embeddings for Language-Guided Retrieval
Chen, Y., & Bazzani, L. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58542-6_9 · Cited by 74 · see citing works
“Learning Using Privileged Information [33, 32] is originally proposed as a learning paradigm to use additional information only available at training time with the purpose of improving model performance on related tasks [30, 17, 20, 21, 12, 41, 7, 16, 18, 2].”
BackgroundAttention-Based Autism Spectrum Disorder Screening With Privileged Modality
Chen, S., & Zhao, Q. (2019) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2019.00127 · Cited by 72 · see citing works
“Unlike existing multi-modal [21, 34] or learning with privileged modality [10, 16, 20, 23] methods, our framework does not rely on the availability of all modalities during deployment or one-to-one pairwise relationship (e.”
BackgroundMethodPrivileged Features Distillation at Taobao Recommendations
Xu, C., Li, Q., Ge, J., et al. (2020) · ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (KDD) · doi:10.1145/3394486.3403309 · Cited by 66 · see citing works
“More importantly, the no-harm guarantee will very likely be violated since estimating the privileged features might be even more challenging than the original problem [20].”
BackgroundHuman Pose Estimation in Extremely Low-Light Conditions
Lee, S., Rim, J., Jeong, B., Kim, G., Woo, B., Lee, H., Cho, S., & Kwak, S. (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/cvpr52729.2023.00075 · Cited by 51 · see citing works
“[32] developed a new dropout operation controlled by privileged information, and Hoffman et al.”
BackgroundObject DGCNN: 3D Object Detection using Dynamic GraphsPreprint
Wang, Y., & Solomon, J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.06923 · Cited by 50 · see citing works
“To mediate uncertainty and improve training efficiency, [58] makes the variance of Dropout [59] a function of privileged information.”
BackgroundHard pixel mining for depth privileged semantic segmentation
Gu, Z., Niu, L., Zhao, H., & Zhang, L. (2021) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2020.3035231 · Cited by 41 · see citing works
“Recently, privileged information has also been utilized in deep learning methods [52], [53], [54], which aim to distill knowledge from privileged information or use privileged information to control the training process.”
BackgroundExploring Privileged Features for Relation Extraction with Contrastive Student-Teacher Learning
Xiaoyan, Z., Yang, M., Qu, Q., Xu, R., & Li, J. (2022) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2022.3161584 · Cited by 39 · see citing works
“Learning using privileged information (LUPI) paradigm endeavors to exploit extra knowledge which is only available during training [38].”
BackgroundMethodTRADI: Tracking Deep Neural Network Weight Distributions
Franchi, G., Bursuc, A., Aldea, E., Dubuisson, S., & Isabelle Bloch, A. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58520-4_7 · Cited by 39 · see citing works
“MC Dropout is currently the most popular instance of BNNs due to its speed and simplicity, with multiple recent extensions [11, 32, 50].”
BackgroundAlehdaghi, M., Josi, A., Cruz, R. M. O., Shamsolmoali, P., & Granger, É. (2025) · IEEE Transactions on Information Forensics and Security · doi:10.1109/tifs.2025.3541969 · Cited by 39 · see citing works
“With LUPI [65], the DL model leverages privileged information (PI) exclusively during the training phase, while the main information remains accessible in both the training and testing phases.”
Robust learning with imperfect privileged information
Li, X., Du, B., Xu, C., Zhang, Y., Zhang, L., & Tao, D. (2020) · Artificial Intelligence · doi:10.1016/j.artint.2020.103246 · Cited by 36 · see citing works
“In convolutional neural networks (CNNs) and recurrent neural networks (RNNs), a heteroscedastic dropout was developed by taking the variance of the dropout as a function of privileged information [26].”
BackgroundVisible-Infrared Person Re-Identification Using Privileged Intermediate Information
Alehdaghi, M., Josi, A., Cruz, R. M. O., & Granger, É. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-25072-9_48 · Cited by 33 · see citing works
Deo, R. C., Grant, R. H., Webb, A. R., Ghimire, S., Igoe, D. P., Downs, N. J., AL‐Musaylh, M. S., Parisi, A. V., & Soar, J. (2022) · Stochastic Environmental Research and Risk Assessment · doi:10.1007/s00477-022-02188-0 · Cited by 32 · see citing works
“The dropout aims to select a fraction of the neurons (defined as a real hyperparameter over the range 0–1) at each model iteration and prevent them from retraining (Lambert et al. 2018; Zhang et al. 2018a; Cai et al. 2019).”
BackgroundLearning with privileged information for photo aesthetic assessment
Shu, Y., Li, Q., Liu, S., & Xu, G. (2020) · Neurocomputing · doi:10.1016/j.neucom.2020.04.142 · Cited by 29 · see citing works
“Lambert et al. [28] propose a new Learning Under Privileged Information algorithm to use a heteroscedastic dropout and make the variance of the dropout a function of privileged information.”
MethodML-DSVM+: A meta-learning based deep SVM+ for computer-aided diagnosis
Han, X., Wang, J., Ying, S., Shi, J., & Shen, D. G. (2022) · Pattern Recognition · doi:10.1016/j.patcog.2022.109076 · Cited by 27 · see citing works
Chen, T., Guo, S., Deng, X., et al. (2024) · Signal Processing · doi:10.1016/j.sigpro.2024.109391 · Cited by 25 · see citing works
Improving Unimodal Object Recognition with Multimodal Contrastive Learning
Meyer, J., Eitel, A., Brox, T., & Burgard, W. (2020) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros45743.2020.9341029 · Cited by 22 · see citing works
Learning to see colours: Biologically relevant virtual staining for adipocyte cell images
Wieslander, H., Gupta, A., Bergman, E., Hallström, E., & Harrison, P. J. (2021) · PLoS ONE · doi:10.1371/journal.pone.0258546 · Cited by 22 · see citing works
Skeleton-Based Human Motion Prediction With Privileged Supervision
Dong, M., & Xu, C. (2022) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2022.3166861 · Cited by 22 · see citing works
“Lambert et al. [61] proposed to apply privileged learning for CNN and recurrent neural network, which uses a heteroscedastic dropout with varying variance.”
KOC+: Kernel ridge regression based one-class classification using privileged information
Gautam, C., Tiwari, A., & Tanveer, M. (2019) · Information Sciences · doi:10.1016/j.ins.2019.07.052 · Cited by 21 · see citing works
Retaining Privileged Information for Multi-Task Learning
Tang, F., Xiao, C., Wang, F., Zhou, J., & Lehman, L.-W. H. (2019) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3292500.3330907 · Cited by 20 · see citing works
“Although recent works such as [15] has generalized the LUPI framework to deep learning settings, our work extends LUPI to allow for multi-task and transfer learning, enabling the generalization of a PI source to accelerate the sample efficiency of many tasks.”
BackgroundMake Me a BNN: A Simple Strategy for Estimating Bayesian Uncertainty from Pre-trained Models
Franchi, G., Laurent, O., Leguéry, M., Bursuc, A., Pilzer, A., & Yao, A. (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/cvpr52733.2024.01159 · Cited by 20 · see citing works
Trust, but Verify: Cross-Modality Fusion for HD Map Change DetectionPreprint
Lambert, J., & Hays, J. H. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2212.07312 · Cited by 20 · see citing works
“We utilize a two-stream architecture [20, 61] with shared parameters, which has been shown to still be effective even for multi-modal input [30].”
MethodSelf-citationPAU-Net: Privileged Action Unit Network for Facial Expression Recognition
Wang, X., Zhang, T., & Chen, C. L. P. (2022) · IEEE Transactions on Cognitive and Developmental Systems · doi:10.1109/tcds.2022.3203822 · Cited by 20 · see citing works
“[39] introduced a special LUPI algorithm for CNN and RNN, which uses PI to control the variance of dropout.”
MethodMutual information-based dropout: Learning deep relevant feature representation architectures
Chen, J., Wu, Z., Zhang, J., et al. (2019) · Neurocomputing · doi:10.1016/j.neucom.2019.04.090 · Cited by 16 · see citing works
Discriminative Multi-View Privileged Information Learning for Image Re-Ranking
Li, J., Xu, C., Yang, W., et al. (2020) · IEEE Transactions on Image Processing · doi:10.1109/tip.2019.2962667 · Cited by 16 · see citing works
“Besides, PI is also embedded into the deep Convolutional Neural Networks (CNNs) and Recurrent Neural Network (RNNs) for image classification and action recognition [19], [20], [24].”
BackgroundLR-SVM+: Learning Using Privileged Information with Noisy Labels
Wu, Z., Xia, X., Wang, R., et al. (2021) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2021.3116417 · Cited by 15 · see citing works
Understanding Dropout for Graph Neural Networks
Shu, J., Xi, B., Li, Y., et al. (2022) · The Web Conference (WWW) · doi:10.1145/3487553.3524725 · Cited by 15 · see citing works
“This framework is similar with[29], they use other side information to control the variance of learnt representations and we use the quantified uncertainty from the model to control the variance.”
MethodInfluentialV-SVR+: Support Vector Regression With Variational Privileged Information
Shu, Y., Li, Q., Xu, C., et al. (2021) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2021.3060955 · Cited by 15 · see citing works
Zhang, H., Guo, L., Li, J., Wang, J., Ying, S., & Shi, J. (2024) · Information Processing & Management · doi:10.1016/j.ipm.2024.103855 · Cited by 12 · see citing works
Liang, J., Jiang, L., & Hauptmann, A. G. (2020) · Cited by 12 · see citing works
IMPROVEMENT OF INCEPTIONV3 MODEL CLASSIFICATION PERFORMANCE USING CHEST X-RAY IMAGES
Kim, J.-H. (2022) · Journal of Mechanics in Medicine and Biology · doi:10.1142/s0219519422400322 · Cited by 12 · see citing works
Audio Representation Learning by Distilling Video as Privileged Information
Hajavi, A., & Etemad, A. (2023) · IEEE Transactions on Artificial Intelligence · doi:10.1109/tai.2023.3243596 · Cited by 11 · see citing works
“Different approaches have been taken in the literature for learning with the help of privileged information [6], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27].”
BackgroundNarykov, O., Johnson, N. T., & Korkin, D. (2021) · Cell Reports · doi:10.1016/j.celrep.2021.110045 · Cited by 11 · see citing works
Gautam, C., Tiwari, A., & Tanveer, M. (2020) · Cognitive Computation · doi:10.1007/s12559-019-09705-4 · Cited by 11 · see citing works
“Most recently, deep learning with privileged information has been explored for multi-class classification task [3].”
BackgroundRobust Asymmetric Learning in POMDPsPreprint
Warrington, A., Lavington, J. W., Ścibior, A., Schmidt, M., & Wood, F. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2012.15566 · Cited by 11 · see citing works
“The notion of using fully observed states unavailable at deployment time is often referred to as exploiting “privileged information” (Vapnik & Vashist, 2009; Lambert et al., 2018).”
B-mode ultrasound based CAD for liver cancers via multi-view privileged information learning
Han, X., Gong, B., Guo, L., Wang, J., Ying, S., Li, S., & Shi, J. (2023) · Neural Networks · doi:10.1016/j.neunet.2023.03.028 · Cited by 10 · see citing works
HalluciDet: Hallucinating RGB Modality for Person Detection Through Privileged Information
Medeiros, H. R., Peña, F. A. G., Aminbeidokhti, M., Dubail, T., Granger, É., & Pedersoli, M. (2024) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv57701.2024.00147 · Cited by 10 · see citing works
“In a classification scenario, additional localization information is used, and its results show that it improves the generalization, requiring fewer samples for the learning process [17].”
BackgroundMurugesan, M., Dhivya, P., Kanna, P. R., & Kumar, G. S. (2023) · Signal Image and Video Processing · doi:10.1007/s11760-023-02847-x · Cited by 9 · see citing works
Gao, Z., Cheng, S., Wittrup, E. M., Gryak, J., & Najarian, K. (2024) · Artificial Intelligence in Medicine · doi:10.1016/j.artmed.2024.102947 · Cited by 9 · see citing works
Wang, H., Wang, G., Li, G., & Lin, L. (2019) · ACM International Conference on Information and Knowledge Management (CIKM) · doi:10.1145/3357384.3357999 · Cited by 9 · see citing works
“Most of the non-structured methods focus on the optimization based on Bayesian learning framework following its variational interpretation[6, 7, 17, 20, 25] or estimate hyperparameters from a prior specific distribution[18, 41, 46].”
BackgroundMulti-Frame to Single-Frame: Knowledge Distillation for 3D Object DetectionPreprint
Wang, Y., Fathi, A., Wu, J., et al. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2009.11859 · Cited by 9 · see citing works
“In contrast to existing works [11,4], which use additional image information and require extra annotations, our model takes advantage of the privileged information that is naturally available in a sequence of point clouds.”
Learning with Privileged Knowledge Distillation for Improved Peptide–Protein Docking
Zhang, Z., Verburgt, J. C., Kagaya, Y., Christoffer, C. W., & Kihara, D. (2025) · ACS Omega · doi:10.1021/acsomega.5c00967 · Cited by 8 · see citing works
Privileged Information Dropout in Reinforcement LearningPreprint
Kamienny, P.-A., Arulkumaran, K., Behbahani, F., Böhmer, W., & Whiteson, S. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2005.09220 · Cited by 8 · see citing works
“PI-Dropout (Lambert et al., 2018) is motivated by information bottleneck (IB) theory (Tishby et al., 2000).”
TriDeNT : Triple deep network training for privileged knowledge distillation in histopathology
Farndale, L., Insall, R. H., & Yuan, K. (2025) · Medical Image Analysis · doi:10.1016/j.media.2025.103479 · Cited by 7 · see citing works
Rodríguez, A. C., D’Aronco, S., Schindler, K., & Wegner, J. D. (2023) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2023.3316718 · Cited by 7 · see citing works
“Alsosimulateddatahasbeen interpretedasprivilegedinformation[3],and(heteroscedastic) dropouthasbeenusedasawayofinjecting,attrainingtime, privilegedinformationintothenetwork[4].”
BackgroundInfluentialDepth Privileged Scene Recognition via Dual Attention Hallucination
Chen, J., Niu, L., & Zhang, L. (2021) · IEEE Transactions on Image Processing · doi:10.1109/tip.2021.3122955 · Cited by 7 · see citing works
Increasing efficiency of SVMp+ for handling missing values in healthcare prediction
Zhang, Y., Gao, Z., Wittrup, E. M., Gryak, J., & Najarian, K. (2023) · PLOS Digital Health · doi:10.1371/journal.pdig.0000281 · Cited by 7 · see citing works
“Apart from SVM, the LUPI paradigm can be applied to other machine learning algorithms and has already been applied to K-means [18] and Convolutional Neural Network for different tasks [13].”
MethodMood Detection from Physical and Neurophysical Data Using Deep Learning Models
Kilimci, Z. H., Güven, A., Uysal, M., & Akyokuş, S. (2019) · Complexity · doi:10.1155/2019/6434578 · Cited by 7 · see citing works
“To alleviate the problem of overfitting, dropout [26] is applied and set to 0.”
BackgroundMethodWieslander, H., Gupta, A., Bergman, E., Hallström, E., & Harrison, P. J. (2021) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2021.01.18.427121 · Cited by 7 · see citing works
Predicting Visual Political Bias Using Webly Supervised Data and an Auxiliary Task
Thomas, C., & Kovashka, A. (2021) · International Journal of Computer Vision · doi:10.1007/s11263-021-01506-3 · Cited by 6 · see citing works
“These approaches use tied weights (Borghi et al., 2018) or multitask training (Elliott and K´ad´ar, 2017), or compute summary statistics (Sharmanska et al., 2013; Lambert et al., 2018), to guide learning.”
BackgroundMethodUsing LUPI to Improve Complex Activity Recognition
Adachi, K., Lago, P., Hattori, Y., & Inoue, S. (2022) · Smart innovation, systems and technologies · doi:10.1007/978-981-19-0361-8_3 · Cited by 6 · see citing works
TRADI: Tracking deep neural network weight distributions for uncertainty estimationPreprint
Franchi, G., Bursuc, A., Aldea, E., Dubuisson, S., & Isabelle Bloch, A. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1912.11316 · Cited by 5 · see citing works
“MC Dropout is currently the most popular instance of BNNs due to its speed and simplicity, with multiple recent extensions [11, 32, 50].”
BackgroundSemi-supervised manifold regularized multi-task learning with privileged information
Liu, B., Li, B., Xiao, Y., et al. (2025) · Information Sciences · doi:10.1016/j.ins.2025.122112 · Cited by 4 · see citing works
A deep machine learning approach for predicting freeway work zone delay using big data
Shabarek, A. (2020) · Cited by 4 · see citing works
Lange, A., Smolyakov, D., & Burnaev, E. V. (2023) · IEEE Access · doi:10.1109/access.2023.3331685 · Cited by 4 · see citing works
“…et al. [40]), domain adaptation (Sarafianos et al. [41]), multiview privileged SVMs (Tang et al. [42]), deep learning under PI (Lambert et al. [43]), PI for structured output prediction (Zhang et al. [44]), label enhancement with multi-label learning (Zhu et al. [45]), PI for the diagnosis of…”
BackgroundTuo, J., Yu, H., Ma, X., & Wang, Y. (2021) · Chinese Control and Decision Conference (CCDC) · doi:10.1109/ccdc52312.2021.9601592 · Cited by 4 · see citing works
“Dropout means that during the training process of a deep learning network, neurons are temporarily dropped from the network with a certain probability p [27].”
BackgroundDeo, R. C., Grant, R. H., Webb, A. R., Ghimire, S., Igoe, D. P., Downs, N. J., AL‐Musaylh, M. S., Parisi, A. V., & Soar, J. (2021) · Research Square · doi:10.21203/rs.3.rs-1069113/v1 · Cited by 4 · see citing works
“The dropout aims to select a fraction of the 397 neurons (defined as a real hyperparameter over the range 0 to 1) at each model iteration and 398 prevent them from retraining [104-106].”
BackgroundSupervision dropout: guidance learning in deep neural network
Zeng, L., Zhang, H., Li, Y., Li, M., & Wang, S. (2022) · Multimedia Tools and Applications · doi:10.1007/s11042-022-14274-0 · Cited by 3 · see citing works
“Utilizing the privileged information of the training data, Heteroscedastic Dropout [21] coaches the deactivation of neurons.”
MethodSplit knowledge transfer in learning under privileged information framework
Gauraha, N., Söderdahl, F., & Spjuth, O. (2019) · Cited by 3 · see citing works
Ienco, D., Gbodjo, Y. J. E., Gaetano, R., & Interdonato, R. (2020) · ISPRS annals of the photogrammetry, remote sensing and spatial information sciences · doi:10.5194/isprs-annals-v-2-2020-997-2020 · Cited by 3 · see citing works
“While such setting is gaining more and more attention in the field of multi-source and multimedia signal processing (Lambert et al., 2018, Chen et al., 2017, Hoffman et al., 2016, Yao et al., 2018), surprisingly, to the best of our knowledge, the only work that makes a connection between the GKD…”
BackgroundRecaspia: Recognizing Carrying Actions in Single Images Using Privileged Information
Smailis, C. V., Vrigkas, M., & Kakadiaris, I. A. (2019) · Proceedings - International Conference on Image Processing · doi:10.1109/icip.2019.8802916 · Cited by 3 · see citing works
“Index Terms— Action Recognition, Static Images, Privileged Information, LUPI, Deep Learning”
MethodInfluentialGautam, C., Tiwari, A., & Tanveer, M. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1904.08338 · Cited by 3 · see citing works
“In recent years, learning using privileged information (LUPI) framework is being quite popular among researchers [1, 2, 3, 4, 5, 6, 7, 8, 9, 10].”
BackgroundHard Pixels Mining: Learning Using Privileged Information for Semantic Segmentation.Preprint
Gu, Z., Niu, L., & Zhang, L. (2019) · arXiv (Cornell University) · Cited by 3 · see citing works
Ding, Z., Wang, Z., Zhang, Y. A., et al. (2024) · Optics and Lasers in Engineering · doi:10.1016/j.optlaseng.2024.108485 · Cited by 2 · see citing works
Demand Forecasting in the Presence of Privileged Information
Ariannezhad, M., Schelter, S., & de Rijke, M. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-65742-0_4 · Cited by 2 · see citing works
“Most of the work done in this area is in the field of computer vision [9,12,17], with teacher-student networks as the dominant approach.”
Neglected Free Lunch – Learning Image Classifiers Using Annotation Byproducts
Han, D., Choe, J., Chun, S.-Y., Chung, J. J. Y., Chang, M., Yun, S., Song, J. Y., & Oh, S. J. (2023) · IEEE International Conference on Computer Vision · doi:10.1109/iccv51070.2023.01848 · Cited by 5 · see citing works
“PI may also be used as a representational bottleneck that regularises the cues for recognition [13, 50, 49].”
BackgroundZhang, L., Bronik, K., Piechnik, S. K., Lima, J. A. C., Neubauer, S., Petersen, S. E., & Frangi, A. F. (2024) · IEEE Transactions on Artificial Intelligence · doi:10.1109/tai.2024.3394798 · Cited by 1 · see citing works
“For clarity, we compared the metric classification with CNN (MC+CNN) in [4], MC+CNN* in [41], and the GTSRB architecture in [30] (MCDA+CNN) with our MCDA+CNN*. Table III shows the accuracy of adaptation and nonadaptation detection for traditional CNN and CNN with PI.”
MethodPremium Access to Convolutional Neural Networks
Bringer, J., Chabanne, H., & Guiga, L. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-68887-5_13 · Cited by 1 · see citing works
Privileged Features Distillation for E-Commerce RecommendationsPreprint
Chen, X., Li, Q., Ge, J., Gao, J., Yang, X., Pei, C., Sun, H., & Ou, W. (2019) · arXiv (Cornell University) · Cited by 1 · see citing works
Privileged Pooling: Better Sample Efficiency Through Supervised AttentionPreprint
Rodríguez, A. C., D’Aronco, S., Schindler, K., & Wegner, J. D. (2020) · arXiv (Cornell University) · Cited by 1 · see citing works
Meta-learning using privileged information for dynamicsPreprint
Day, B., Norcliffe, A., Moss, J., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2104.14290 · Cited by 1 · see citing works
“Hern´andez-Lobato et al. (2014) show that privileged information can be used effectively by GP classifiers, whilst Lambert et al. (2018) apply LUPI to deep neural networks by setting the dropout rate to be a function of the privileged information.”
BackgroundImproving the trustworthiness of image classification models by utilizing bounding-box annotationsPreprint
Kc, D., & Zhang, C. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2108.10131 · Cited by 1 · see citing works
“There have been considerable efforts addressing the above question, with most progress focusing on improving test accuracy (e.g. [30, 11, 13]).”
BackgroundTGN-PL: Learning to Socialize Using Privileged Information and Temporal Graph Networks
Chew, J. Y., Kim, J. T., & Ha, S. (2025) · doi:10.1145/3747327.3763033 · Cited by 1 · see citing works
“However, previous LUPI instances have focused on image classification [8] or single-agent reinforcement learning [13] and have not yet addressed dynamic social graphs under real-time constraints.”
Privileged Pooling: Supervised attention-based pooling for compensating dataset bias
Rodríguez, A. C., D’Aronco, S., Schindler, K., & Wegner, J. D. (2020)
Heterogeneous Representation Learning: A ReviewPreprint
Zhou, J. T., Peng, X., & Ong, Y.-S. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2004.13303
Automatic Annotation of Change in Earth Observation Imagery.
Neptune, N. (2020)
Ferrari, C., Berretti, S., & Bimbo, A. D. (2021) · Proceedings - International Conference on Pattern Recognition/Proceedings/International Conference on Pattern Recognition · doi:10.1109/icpr48806.2021.9412994
“Lambert et al. [17] proposed to use a heteroscedastic dropout by making the variance of the dropout a function of privileged information.”
BackgroundSacco, L., Ienco, D., & Interdonato, R. (2021) · Agritrop (Cirad)
Foreground-Specialized Model Imitation for Instance Segmentation
Li, D., Li, W., & Jin, H. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-26293-7_24
Through the Eyes of the Expert: Aligning Human and Machine Attention for Industrial AI
Koebler, A., Greisinger, C., Paulus, J., et al. (2024) · Lecture notes in computer science · doi:10.1007/978-3-031-60611-3_28
Gate-ViT: Gated Vision Transformer for Fine-Grained Visual Classification
Lu, X., Wang, K., Wang, P., Zhang, Q., Zhao, Y., Liu, G., & Yu, X. (2025) · Lecture notes in computer science · doi:10.1007/978-981-96-8180-8_37
Deep-multiscale stratified aggregation
Wu, Z., Song, Y., Hu, F., et al. (2025) · Signal Image and Video Processing · doi:10.1007/s11760-025-04820-2
“ELAN differs in that it cascades and aggregates information from different layers during successive convolutions to bet - ter preserve multiscale information, but this approach loses features from previous layers appearing as an information bottleneck [11].”
O’Brien, G., Tu, Z., Ding, Q., et al. (2025) · doi:10.1109/embc58623.2025.11252678
“FCN trained with LUPI framework using heteroscedastic dropout [15].”
Latent Behavior Cloning for Deep Reinforcement Learning Robotic Tasks
Voulgarakis, R., Katsikas, D., Passalis, N., & Tefas, A. (2025) · European Signal Processing Conference (EUSIPCO) · doi:10.23919/eusipco63237.2025.11226088
“…approach to merging expert supervision with reinforcement learning. b) Learning Using Privileged Information: Learning Using Privileged Information (LUPI) was introduced by Vapnik and Vashist for support vector machines [25] and has since been adopted in supervised learning settings [13], [19].”
Pérez, O. M., Diers, B. W., & Martin, N. F. (2026) · Plant Phenomics · doi:10.1016/j.plaphe.2026.100189
“An improvement that could be achieved by learning latent relationships, reducing noise, or clarifying decision boundaries [39,40], thereby enhancing how RF chooses splits when classifying gray, light tawny, and tawny pubescence.”
- Authors
- Lambert, J., Sener, O., & Savarese, S.
- Affiliations
- Georgia Institute of Technology (US)
- Stanford University (US)
- Venue
- 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition
- Publisher
- IEEE
- Published
- 2018-06
- Pages
- 8886-8895
- ISSN
- 1063-6919 (ISSN-L)
- OpenAlex
- W2798520291
- Semantic Scholar
- 4571aa4e5cc335fd71f4fdce13c9a80419b90aaa
- Cited by
- 93
- References
- 41
- Citations by year
- 2026: 1 · 2025: 9 · 2024: 10 · 2023: 10 · 2022: 10 · 2021: 17 · 2020: 22 · 2019: 14
- FWCI
- 5.27
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- Multidisciplinary University Research Initiative
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning and Algorithms
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning and Data Classification
- Keywords
- privileged information, uncertainty modeling, sample efficiency, convolutional neural networks, recurrent neural networks, image classification, machine translation, generalization bounds
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.
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
