Estimating Example Difficulty using Variance of Gradients
Open accessThe DOI matches its Crossref record; 30 of the 30 references it cites were found in a registry or an index.
Abstract
In machine learning, a question of great interest is understanding what examples are challenging for a model to classify. Identifying atypical examples ensures the safe de-ployment of models, isolates samples that require further human inspection and provides interpretability into model behavior. In this work, we propose Variance of Gradients (VoG) as a valuable and efficient metric to rank data by difficulty and to surface a tractable subset of the most chal-lenging examples for human-in-the-loop auditing. We show that data points with high VoG scores are far more difficult for the model to learn and over-index on corrupted or mem-orized examples. Further, restricting the evaluation to the test set instances with the lowest VoG improves the model's generalization performance. Finally, we show that VoG is a valuable and efficient ranking for out-of-distribution detection.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
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
Zagoruyko, S., & Komodakis, N. (2016) · Procedings of the British Machine Vision Conference 2016 · doi:10.5244/c.30.87 · Cited by 7,244
Bengio, Y., Louradour, J., Collobert, R., & Weston, J. (2009) · Proceedings of the 26th Annual International Conference on Machine Learning · doi:10.1145/1553374.1553380 · Cited by 5,366 · cited 1×
“Given the size of current datasets, this can be a powerful interpretability tool to isolate a tractable subset of examples for human-in-the-loop auditing and aid in curriculum learning [8] or distinguishing between sources of uncertainty [16, 33].”
BackgroundTHE GENERALIZATION OF ‘STUDENT'S’ PROBLEM WHEN SEVERAL DIFFERENT POPULATION VARLANCES ARE INVOLVED
WELCH, B. L. (1947) · Biometrika · doi:10.1093/biomet/34.1-2.28 · Cited by 4,135
Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles
Lakshminarayanan, B., Pritzel, A., & Blundell, C. (2016) · arXiv · doi:10.48550/arxiv.1612.01474 · Cited by 3,666 · cited 1×
“Recent works have highlighted that DNNs produce un-calibrated output probabilities that cannot be interpreted as a measure of certainty [22, 26, 37, 44].”
BackgroundOn calibration of modern neural networks
Guo, C., Pleiss, G., Sun, Y., & Weinberger, K. Q. (2017) · arXiv · doi:10.48550/arxiv.1706.04599 · Cited by 3,435
What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision?
Kendall, A., & Gal, Y. (2017) · arXiv · doi:10.48550/arxiv.1703.04977 · Cited by 3,088
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 · cited 2×
“To do this, we replicate the general experiment setup of Zhang et al. [72] and replace 20% of all labels in the training set with randomly shuffled labels.”
BackgroundMethodLearning Important Features Through Propagating Activation Differences
Shrikumar, A., Greenside, P., & Kundaje, A. (2017) · arXiv · doi:10.48550/arxiv.1704.02685 · Cited by 2,761
Unsupervised Anomaly Detection with Generative Adversarial Networks to Guide Marker Discovery
Schlegl, T., Seeböck, P., Waldstein, S. M., Schmidt-Erfurth, U., & Langs, G. (2017) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-59050-9_12 · Cited by 2,627
A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attacks
Lee, K., Lee, K., Lee, H., & Shin, J. (2018) · arXiv · doi:10.48550/arxiv.1807.03888 · Cited by 2,450
A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks
Hendrycks, D., & Gimpel, K. (2016) · arXiv · doi:10.48550/arxiv.1610.02136 · Cited by 1,968
Understanding Black-box Predictions via Influence Functions
Koh, P. W., & Liang, P. (2017) · arXiv · doi:10.48550/arxiv.1703.04730 · Cited by 1,514 · cited 1×
“In addition, VoG is far more efficient than other global rankings like [42] and [23].”
BackgroundSmoothGrad: removing noise by adding noise
Smilkov, D., Thorat, N., Kim, B., Viégas, F., & Wattenberg, M. (2017) · arXiv · doi:10.48550/arxiv.1706.03825 · Cited by 1,467
Hendrycks, D., Zhao, K., Basart, S., Steinhardt, J., & Song, D. (2021) · 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr46437.2021.01501 · Cited by 1,090
Hendrycks, D., Zhao, K., Basart, S., Steinhardt, J., & Song, D. (2019) · arXiv · doi:10.48550/arxiv.1907.07174 · Cited by 1,058
Li, O., Liu, H., Chen, C., & Rudin, C. (2018) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v32i1.11771 · Cited by 541
Veale, M., & Binns, R. (2017) · Big Data & Society · doi:10.1177/2053951717743530 · Cited by 512
Veale, M., & Binns, R. (2017) · doi:10.31235/osf.io/ustxg · Cited by 512
Leveraging uncertainty information from deep neural networks for disease detection
Leibig, C., Allken, V., Ayhan, M. S., Berens, P., & Wahl, S. (2017) · Scientific Reports · doi:10.1038/s41598-017-17876-z · Cited by 510 · cited 1×
“…through case-based reasoning [11,30,39], it can also be used to surface a tractable subset of atypical examples for further human auditing [46,73], for active learning to inform model improvements, and to choose not to clas-* sify some instances when the model is uncertain [7,14,21].”
BackgroundHidden stratification causes clinically meaningful failures in machine learning for medical imaging
Oakden-Rayner, L., Dunnmon, J., Carneiro, G., & Re, C. (2020) · Proceedings of the ACM Conference on Health, Inference, and Learning · doi:10.1145/3368555.3384468 · Cited by 411 · cited 1×
“Over the past decade, machine learning models are increasingly deployed to high-stake decision applications such as healthcare [4, 20, 52, 70], self-driving cars [51] and finance [53].”
BackgroundDeep Semi-Supervised Anomaly Detection
Ruff, L., Vandermeulen, R. A., Görnitz, N., Binder, A., Müller, E., Müller, K.-R., & Kloft, M. (2019) · arXiv · doi:10.48550/arxiv.1906.02694 · Cited by 388
Deep learning predicts hip fracture using confounding patient and healthcare variables
Badgeley, M. A., Zech, J. R., Oakden-Rayner, L., Glicksberg, B. S., Liu, M., Gale, W., McConnell, M. V., Percha, B., Snyder, T. M., & Dudley, J. T. (2019) · npj Digital Medicine · doi:10.1038/s41746-019-0105-1 · Cited by 290
Does learning require memorization? a short tale about a long tail
Feldman, V. (2020) · Proceedings of the 52nd Annual ACM SIGACT Symposium on Theory of Computing · doi:10.1145/3357713.3384290 · Cited by 287 · cited 1×
“Overparameterized networks have been shown to achieve zero training error by memorizing examples [19,32, 72].”
BackgroundHooker, S. (2021) · Communications of the ACM · doi:10.1145/3467017 · Cited by 211
Self-citationHooker, S. (2020) · arXiv · doi:10.48550/arxiv.2009.06489 · Cited by 185
Self-citationMoving beyond “algorithmic bias is a data problem”
Hooker, S. (2021) · Patterns · doi:10.1016/j.patter.2021.100241 · Cited by 181 · cited 1×
“Even acquiring the labels for a limited number of attributes protected by law (gender, race) is expensive and/or may be perceived as intrusive, leading to noisy or incomplete labels [2,29].”
BackgroundSelf-citationAndrus, M., Spitzer, E., Brown, J., & Xiang, A. (2021) · Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency · doi:10.1145/3442188.3445888 · Cited by 111
Guha Roy, A., Ren, J., Azizi, S., Loh, A., Natarajan, V., Mustafa, B., Pawlowski, N., Freyberg, J., Liu, Y., Beaver, Z., et al. (2022) · Medical Image Analysis · doi:10.1016/j.media.2021.102274 · Cited by 77 · cited 1×
“Besides aiding interpretability through case-based reasoning [11, 30, 39], it can also be used to surface a tractable subset of atypical examples for further human auditing [46, 73], for active learning to inform model improvements, and to choose not to classify some instances when the model is uncertain [7, 14, 21].”
MethodHow to Explain Individual Classification Decisions
(2010) · Journal of Machine Learning Research · doi:10.5555/1756006.1859912 · Cited by 54
Cited by
HIVE: Evaluating the Human Interpretability of Visual Explanations
Kim, S. S. Y., Meister, N., Ramaswamy, V. V., Fong, R., & Russakovsky, O. (2022) · Lecture Notes in Computer Science · doi:10.1007/978-3-031-19775-8_17 · Cited by 72
Generating High Fidelity Data from Low-density Regions using Diffusion Models
Sehwag, V., Hazirbas, C., Gordo, A., Ozgenel, F., & Ferrer, C. C. (2022) · 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52688.2022.01120 · Cited by 41
“Samples from low-density regions often comprise novel attributes (Figure 1a) and have higher entropy than high-density samples [1].”
BackgroundCombining Ensembles and Data Augmentation can Harm your Calibration
Wen, Y., Jerfel, G., Muller, R., Dusenberry, M. W., Snoek, J., Lakshminarayanan, B., & Tran, D. (2020) · arXiv · doi:10.48550/arxiv.2010.09875 · Cited by 27
Metadata
- Authors
- Agarwal, C., D'souza, D., & Hooker, S.
- Venue
- 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
- Publisher
- IEEE
- Published
- 2022-06
- Pages
- 10358-10368
- ISSN
- 1063-6919 (ISSN-L)
- OpenAlex
- W3080532707
- Cited by
- 65
- References
- 30
- FWCI
- 8.09
- Type
- conference paper
- Language
- en
- Open access
- green
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
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.