Finding Label and Model Errors in Perception Data With Learned Observation Assertions
Summary: Introduces learned observation assertions; Fixy audits perception labels. Fixy learns distributions over noisy labels and prior models to score label errors, outperforming baselines with up to 2x precision and uncovering errors in 70% of scenes. (summarized by gpt-5-nano on Feb 09 2026)
Incoming Non-self Citations Over Time
Authors
- 1. Daniel Kang (Stanford University)
- 2. Nikos Arechiga (Toyota)
- 3. Sudeep Pillai (Toyota)
- 4. Peter D. Bailis (Stanford University)
- 5. Matei Zaharia (Stanford University)
BibTeX Citation
@inproceedings{kang_sigmod22,
title = {{Finding Label and Model Errors in Perception Data With Learned Observation Assertions}},
author = {Kang, Daniel and Arechiga, Nikos and Pillai, Sudeep and Bailis, Peter D. and Zaharia, Matei},
series = {{SIGMOD} '22},
booktitle = {Proceedings of the {ACM} {SIGMOD} International Conference on Management of Data},
publisher = {Association for Computing Machinery},
doi = {10.1145/3514221.3517907},
url = {https://dl.acm.org/doi/10.1145/3514221.3517907},
year = {2022}
}Incoming Citations (Sorted by Pagerank)
Showing 2 of 2 citing papers.
| Rank | Citing Paper | Year | Venue | Pagerank |
|---|---|---|---|---|
| 4,621 | spade: Synthesizing Data Quality Assertions for Large Language Model Pipelines | 2024 | VLDB | 6.6024412e-05 |
| 6,443 | AutoCTS+: Joint Neural Architecture and Hyperparameter Search for Correlated Time Series Forecasting | 2023 | SIGMOD | 5.8775356e-05 |
Previous
Page 1 / 1
Next
Outgoing Citations (Sorted by Pagerank)
Showing 8 of 8 cited papers.
Citations counted here include only citations to other VLDB/SIGMOD/CIDR/PODS papers in this database.
| Rank | Cited Paper | Year | Venue | Pagerank |
|---|---|---|---|---|
| 112 | HoloClean: Holistic Data Repairs with Probabilistic Inference | 2017 | VLDB | 0.00032801121 |
| 185 | A Cost-Based Model and Effective Heuristic for Repairing Constraints by Value Modification | 2005 | SIGMOD | 0.00026231189 |
| 205 | Snorkel: Rapid Training Data Creation with Weak Supervision | 2018 | VLDB | 0.00025235185 |
| 946 | HoloDetect: Few-Shot Learning for Error Detection | 2019 | SIGMOD | 0.00013054126 |
| 1,147 | Data Management Challenges in Production Machine Learning | 2017 | SIGMOD | 0.00011974846 |
| 1,323 | Data Cleaning: Overview and Emerging Challenges | 2016 | SIGMOD | 0.00011152602 |
| 1,332 | Sampling the Repairs of Functional Dependency Violations under Hard Constraints | 2010 | VLDB | 0.00011129529 |
| 11,824 | Leveraging Organizational Resources to Adapt Models to New Data Modalities | 2020 | VLDB | 5.093636e-05 |
Previous
Page 1 / 1
Next
Semantically Similar Papers
| # | Overall Rank | Paper | Year | Venue |
|---|---|---|---|---|
| 1 | 3,376 | Robust and Transferable Log-based Anomaly Detection | 2023 | SIGMOD |
| 2 | 10,785 | Auto-Test: Learning Semantic-Domain Constraints for Unsupervised Error Detection in Tables | 2025 | SIGMOD |
| 3 | 4,959 | QFix: Diagnosing Errors through Query Histories | 2017 | SIGMOD |
| 4 | 8,225 | Deducing Certain Fixes to Graphs | 2019 | VLDB |
| 5 | 9,245 | Towards Observability for Production Machine Learning Pipelines | 2022 | VLDB |
| 6 | 946 | HoloDetect: Few-Shot Learning for Error Detection | 2019 | SIGMOD |
| 7 | 2,981 | Towards Dependable Data Repairing with Fixing Rules | 2014 | SIGMOD |
| 8 | 11,211 | MisDetect: Iterative Mislabel Detection using Early Loss | 2024 | VLDB |
| 9 | 8,594 | Exploratory Training: When Annotators Learn About Data | 2023 | SIGMOD |
| 10 | 3,670 | Learning to Validate the Predictions of Black Box Classifiers on Unseen Data | 2020 | SIGMOD |