Adversarial Learning for Feature Shift Detection and Correction
Not on the map. Why?
Abstract (arXiv)
Data shift is a phenomenon present in many real-world applications, and while there are multiple methods attempting to detect shifts, the task of localizing and correcting the features originating such shifts has not been studied in depth. Feature shifts can occur in many datasets, including in multi-sensor data, where some sensors are malfunctioning, or in tabular and structured data, including biomedical, financial, and survey data, where faulty standardization and data processing pipelines can lead to erroneous features. In this work, we explore using the principles of adversarial learning, where the information from several discriminators trained to distinguish between two distributions is used to both detect the corrupted features and fix them in order to remove the distribution shift between datasets. We show that mainstream supervised classifiers, such as random forest or gradient boosting trees, combined with simple iterative heuristics, can localize and correct feature shifts, outperforming current statistical and neural network-based techniques. The code is available at https://github.com/AI-sandbox/DataFix.
Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Builds on 30 of its 103 references that are in the map. First posted on arXiv 7 Dec 2023.
Builds on
30 of 103 references.
References · 30
- 2015Domain-Adversarial Training of Neural Networkscited by 3,023
- 2017ZOO: Zeroth Order Optimization based Black-box Attacks to Deep Neural Networks without Training Substitute Modelscited by 656
- 2014OpenML: networked science in machine learningcited by 523
- 2016f-GAN: Training Generative Neural Samplers using Variational Divergence Minimizationcited by 504
- 2020Learning under Concept Drift: A Reviewcited by 470
- 2018GAIN: Missing Data Imputation using Generative Adversarial Netscited by 349
- 2018Failing Loudly: An Empirical Study of Methods for Detecting Dataset Shiftcited by 240
- 2018CatBoost: gradient boosting with categorical features supportcited by 208
- 2023Data-centric Artificial Intelligence: A Surveycited by 202
- 2019Generalizing to unseen domains via distribution matchingcited by 109
- 2014Matrix Completion and Low-Rank SVD via Fast Alternating Least Squarescited by 104
- 2020Missing Data Imputation using Optimal Transportcited by 99
18 more
- 2022HyperImpute: Generalized Iterative Imputation with Automatic Model Selectioncited by 81
- 2019MisGAN: Learning from Incomplete Data with Generative Adversarial Networkscited by 74
- 2019Deep Weakly-supervised Anomaly Detectioncited by 59
- 2020Interactive Weak Supervision: Learning Useful Heuristics for Data Labelingcited by 52
- 2021MIRACLE: Causally-Aware Imputation via Learning Missing Data Mechanismscited by 50
- 2020Dataset Discovery in Data Lakescited by 31
- 2021Feature Shift Detection: Localizing Which Features Have Shifted via Conditional Distribution Testscited by 24
- 2019AlphaClean: Automatic Generation of Data Cleaning Pipelinescited by 23
- 2020PC-GAIN: Pseudo-label Conditional Generative Adversarial Imputation Networks for Incomplete Datacited by 15
- 2021Multiple Imputation via Generative Adversarial Network for High-dimensional Blockwise Missing Value Problemscited by 15
- 2021MLDemon: Deployment Monitoring for Machine Learning Systemscited by 14
- 2018Request-and-Reverify: Hierarchical Hypothesis Testing for Concept Drift Detection with Expensive Labelscited by 11
- 2016Online Decorrelation of Humidity and Temperature in Chemical Sensors for Continuous Monitoringcited by 8
- 2020Diverse Instances-Weighting Ensemble based on Region Drift Disagreement for Concept Drift Adaptationcited by 8
- 2021Iterative Alignment Flowscited by 4
- 2021Filter Methods for Feature Selection in Supervised Machine Learning Applications -- Review and Benchmarkcited by 4
- 2017Parametric Adversarial Divergences are Good Losses for Generative Modelingcited by 1
- 2018Convergence Rates for Empirical Estimation of Binary Classification Boundscited by 1
Built on by
Citing papers per year: 2025: 2.
2 papers.