Losses over Labels: Weakly Supervised Learning via Direct Loss Construction
Open accessAbstract
Owing to the prohibitive costs of generating large amounts of labeled data, programmatic weak supervision is a growing paradigm within machine learning. In this setting, users design heuristics that provide noisy labels for subsets of the data. These weak labels are combined (typically via a graphical model) to form pseudolabels, which are then used to train a downstream model. In this work, we question a foundational premise of the typical weakly supervised learning pipeline: given that the heuristic provides all “label” information, why do we need to generate pseudolabels at all? Instead, we propose to directly transform the heuristics themselves into corresponding loss functions that penalize differences between our model and the heuristic. By constructing losses directly from the heuristics, we can incorporate more information than is used in the standard weakly supervised pipeline, such as how the heuristics make their decisions, which explicitly informs feature selection during training. We call our method Losses over Labels (LoL) as it creates losses directly from heuristics without going through the intermediate step of a label. We show that LoL improves upon existing weak supervision methods on several benchmark text and image classification tasks and further demonstrate that incorporating gradient information leads to better performance on almost every task.
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?
Recent advancements in NLU prominently feature large transformer-based models like BERT and GPT-3, which have significantly advanced the field by enabling deeper and more complex understanding of language subtleties.
“Recent advancements have prominently featured large transformer-based models like BERT and GPT-3, which have significantly advanced the field by enabling a deeper and more complex understanding of language subtleties [197], [198].”
From From Google Gemini to OpenAI Q* (Q-Star): A Survey on Reshaping the Generative Artificial Intelligence (AI) Research Landscape (2025) · cites this paper as [197]
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
Zero-Shot Learning—A Comprehensive Evaluation of the Good, the Bad and the Ugly
Xian, Y., Lampert, C. H., Schiele, B., & Akata, Z. (2019) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2018.2857768 · Cited by 1,678 · see citing works
Maximum Likelihood Estimation of Observer Error-Rates Using the EM Algorithm
Dawid, A. P., & Skene, A. M. (1979) · Applied Statistics · doi:10.2307/2346806 · Cited by 1,623 · see citing works
Making Deep Neural Networks Robust to Label Noise: A Loss Correction Approach
Patrini, G., Rozza, A., Menon, A. K., Nock, R., & Qu, L. (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.2017.240 · Cited by 1,464 · see citing works
Argyriou, A., Evgeniou, T., & Pontil, M. (2007) · The MIT Press eBooks · doi:10.7551/mitpress/7503.003.0010 · Cited by 1,376 · see citing works
Certified Adversarial Robustness via Randomized SmoothingPreprint
Cohen, J. M., Rosenfeld, E., & Kolter, J. Z. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1902.02918 · Cited by 1,172 · see citing works
Self-citationSnorkel: rapid training data creation with weak supervision
Ratner, A., Bach, S. H., Ehrenberg, H. R., Fries, J., Wu, S., & Ré, C. (2019) · The VLDB Journal · doi:10.1007/s00778-019-00552-1 · Cited by 808 · see citing works
Multitask Learning: A Knowledge-Based Source of Inductive Bias
Caruana, R. (1993) · Elsevier eBooks · doi:10.1016/b978-1-55860-307-3.50012-5 · Cited by 746 · see citing works
Exploiting Task Relatedness for Multiple Task Learning
Ben-David, S., & Schuller, R. (2003) · Lecture notes in computer science · doi:10.1007/978-3-540-45167-9_41 · Cited by 432 · see citing works
Data Programming: Creating Large Training Sets, Quickly
Ratner, A., De, C., Wu, S., Selsam, D., & Ré, C. (2016) · Europe PMC (PubMed Central) · doi:10.48550/arxiv.1605.07723 · Cited by 385 · see citing works
Optimal Decision Rules in Uncertain Dichotomous Choice Situations
Nitzan, S. I., & Paroush, J. (1982) · International Economic Review · doi:10.2307/2526438 · Cited by 303 · see citing works
The strength of weak learnability
Schapire, R. E. (1989) · Annual Symposium on Foundations of Computer Science · doi:10.1109/sfcs.1989.63451 · Cited by 264 · see citing works
Learning from crowdsourced labeled data: a survey
Zhang, J., Wu, X., & Sheng, V. S. (2016) · Artificial Intelligence Review · doi:10.1007/s10462-016-9491-9 · Cited by 203 · see citing works
Zero-Shot Learning -- A Comprehensive Evaluation of the Good, the Bad and the UglyPreprint
Xian, Y., Lampert, C. H., Schiele, B., et al. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1707.00600 · Cited by 171 · see citing works
Boosting a weak learning algorithm by majority
Freund, Y. (1990) · Elsevier eBooks · doi:10.1016/b978-1-55860-146-8.50019-9 · Cited by 167 · see citing works
Aggregating crowdsourced binary ratings
Dalvi, N. N., Dasgupta, A., Kumar, R., et al. (2013) · International World Wide Web Conference (WWW) · doi:10.1145/2488388.2488414 · Cited by 166 · see citing works
Training Complex Models with Multi-Task Weak Supervision
Ratner, A., Hancock, B., Dunnmon, J., et al. (2019) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v33i01.33014763 · Cited by 147 · see citing works
Making Deep Neural Networks Robust to Label Noise: a Loss Correction ApproachPreprint
Patrini, G., Rozza, A., Menon, A. K., et al. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1609.03683 · Cited by 112 · see citing works
Sluice networks: Learning what to share between loosely related tasks.Preprint
Ruder, S., Bingel, J., Augenstein, I., & Søgaard, A. (2017) · arXiv (Cornell University) · Cited by 111 · see citing works
Trace Norm Regularised Deep Multi-Task LearningPreprint
Yang, Y., & Hospedales, T. M. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1606.04038 · Cited by 104 · see citing works
A Survey of Label-noise Representation Learning: Past, Present and FuturePreprint
Han, B., Yao, Q., Liu, T., et al. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.04406 · Cited by 103 · see citing works
Yu, Y., Zuo, S., Jiang, H., Ren, W., Zhao, T., & Zhang, C. (2021) · Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT) · doi:10.18653/v1/2021.naacl-main.84 · Cited by 98 · see citing works
Weakly Supervised Sequence Tagging from Noisy Rules
Safranchik, E., Luo, S., & Bach, S. H. (2020) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v34i04.6009 · Cited by 79 · see citing works
Self-Training with Weak Supervision
Karamanolakis, G., Mukherjee, S., Zheng, G., & Awadallah, A. H. (2021) · Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT) · doi:10.18653/v1/2021.naacl-main.66 · Cited by 67 · see citing works
Minimax Optimal Convergence Rates for Estimating Ground Truth from Crowdsourced LabelsPreprint
Gao, C., & Zhou, D. (2013) · arXiv (Cornell University) · doi:10.48550/arxiv.1310.5764 · Cited by 57 · see citing works
WRENCH: A Comprehensive Benchmark for Weak SupervisionPreprint
Zhang, J., Yu, Y., Li, Y., Wang, Y., Yang, Y., Yang, M., & Ratner, A. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2109.11377 · Cited by 50 · see citing works
Learning from Rules Generalizing Labeled ExemplarsPreprint
Awasthi, A., Ghosh, S., Goyal, R., & Sarawagi, S. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2004.06025 · Cited by 48 · see citing works
Fast and Three-rious: Speeding Up Weak Supervision with Triplet MethodsPreprint
Fu, D. Y., Chen, M. F., Sala, F., Hooper, S., Fatahalian, K., & Ré, C. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2002.11955 · Cited by 42 · see citing works
A Survey on Programmatic Weak SupervisionPreprint
Zhang, J., Hsieh, C.-Y., Yu, Y., et al. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2202.05433 · Cited by 37 · see citing works
Universalizing Weak SupervisionPreprint
Shin, C.-H., Li, W., Vishwakarma, H., Roberts, N., & Sala, F. (2021) · International Conference on Learning Representations · doi:10.48550/arxiv.2112.03865 · Cited by 36 · see citing works
Snorkel DryBell: A Case Study in Deploying Weak Supervision at Industrial ScalePreprint
Bach, S. H., Rodriguez, D., Liu, Y., et al. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1812.00417 · Cited by 24 · see citing works
Curriculum Labeling: Self-paced Pseudo-Labeling for Semi-Supervised Learning.
Cascante-Bonilla, P., Tan, F., Qi, Y., et al. (2020) · Cited by 15 · see citing works
Optimally Combining Classifiers Using Unlabeled DataPreprint
Balsubramani, A., & Freund, Y. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1503.01811 · Cited by 14 · see citing works
Scalable Semi-Supervised Aggregation of ClassifiersPreprint
Balsubramani, A., & Freund, Y. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1506.05790 · Cited by 11 · see citing works
Improving Label Noise Filtering by Exploiting Unlabeled Data
Guan, D., Wei, H., Yuan, W., et al. (2018) · IEEE Access · doi:10.1109/access.2018.2807779 · Cited by 10 · see citing works
DP-SSL: Towards Robust Semi-supervised Learning with A Few Labeled SamplesPreprint
Xu, Y., Ding, J., Zhang, L., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.13740 · Cited by 6 · see citing works
Yu, Y., Zuo, S., Jiang, H., Ren, W., Zhao, T., & Zhang, C. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.07835 · Cited by 5 · see citing works
Semi-Supervised Aggregation of Dependent Weak Supervision Sources With Performance Guarantees
Mazzetto, A., Sam, D., PARK, A. W., et al. (2021) · International Conference on Artificial Intelligence and Statistics · Cited by 5 · see citing works
Self-citationConstrained Labeling for Weakly Supervised LearningPreprint
Arachie, C., & Huang, B. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2009.07360 · Cited by 5 · see citing works
Comparing the Value of Labeled and Unlabeled Data in Method-of-Moments Latent Variable EstimationPreprint
Chen, M. F., Cohen-Wang, B., Mussmann, S., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.02761 · Cited by 3 · see citing works
Creating Training Sets via Weak Indirect SupervisionPreprint
Zhang, J., Wang, B., Song, X., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.03484 · Cited by 3 · see citing works
Stochastic Generalized Adversarial Label LearningPreprint
Arachie, C., & Huang, B. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1906.00512 · Cited by 1 · see citing works
Data Consistency for Weakly Supervised LearningPreprint
Arachie, C., & Huang, B. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2202.03987 · Cited by 1 · see citing works
Active WeaSuL: Improving Weak Supervision with Active LearningPreprint
Biegel, S., El-Khatib, R., Oliveira, L. O. V. B., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2104.14847
Losses over Labels: Weakly Supervised Learning via Direct Loss ConstructionPreprint
Sam, D., & Kolter, J. Z. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2212.06921
Self-citation
Flagged first, then by citations
Research works printed but not linked: 10
- Breiman, L. 1996. Bagging Predictors. Machine Learning, 24(2): 123-140.
- Chen, T.; and Guestrin, C. 2016. XGBoost. Proceed- ings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.
- Karamanolakis, G.; Mukherjee, S. S.; Zheng, G.; and Awadallah, A. H. 2021. Self-training with Weak Supervi- sion. In NAACL 2021.
- Lee, D.-H. 2013. Pseudo-Label: The Simple and Efficient Semi-Supervised Learning Method for Deep Neural Net- works. Workshop on challenges in representation learning, ICML.
- Mazzetto, A.; Cousins, C.; Sam, D.; Bach, S. H.; and Upfal, E. 2021a. Adversarial Multi Class Learning under Weak Supervision with Performance Guarantees. In International Conference on Machine Learning, 7534-7543. PMLR.
- PMLR.
- Natarajan, N.; Dhillon, I. S.; Ravikumar, P. K.; and Tewari, A. 2013. Learning with Noisy Labels. In Burges, C. J. C.; Bottou, L.; Welling, M.; Ghahramani, Z.; and Weinberger, K. Q., eds., Advances in Neural Information Processing Sys- tems, volume 26.
- Pham, H.; Xie, Q.; Dai, Z.; and Le, Q. V. 2021. Meta Pseudo Labels. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 11552-11563.
- Ruhling Cachay, S.; Boecking, B.; and Dubrawski, A. 2021. End-to-End Weak Supervision. Advances in Neural Infor- mation Processing Systems, 34.
- Yu, P.; Ding, T.; and Bach, S. H. 2022. Learning from Mul- tiple Noisy Partial Labelers. In Artificial Intelligence and Statistics (AISTATS).
Cited by
Sorted by citations
McIntosh, T. R., Susnjak, T., Liu, T., Watters, P., Xu, D., Liu, D., & Halgamuge, M. N. (2025) · Technologies · doi:10.3390/technologies13020051 · Cited by 102 · see citing works
- Authors
- Sam, D., & Kolter, J. Z.
- Affiliations
- Carnegie Mellon University (US)
- ORCID
- Kolter, J. Zico
- Venue
- Proceedings of the AAAI Conference on Artificial Intelligence
- Published
- 2023-06-26
- Volume
- 37
- Issue
- 8
- Pages
- 9695-9703
- ISSN
- 2159-5399 (ISSN-L)
- 2374-3468
- OpenAlex
- W4382203110
- Cited by
- 6
- References
- 44
- Citations by year
- 2025: 1 · 2024: 3 · 2023: 2
- FWCI
- 1.07
- Type
- conference paper
- Language
- en
- Open access
- diamond
- dx.doi.org
- Locations
- Proceedings of the AAAI Conference on Artificial Intelligence (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- AAAI Publications (The Association for the Advancement of Artificial Intelligence (AAAI)) (submittedVersion)
- arXiv (Cornell University)
- Funders
- Achievement Rewards for College Scientists Foundation
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning and Data Classification
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning and Algorithms
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Keywords
- weakly supervised learning, label noise, pseudo-labeling, feature selection, gradient information, text classification, image classification
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
