Towards Making Systems Forget with Machine Unlearning
By Yinzhi Cao, Junfeng Yang. 2015
1,123 citations recorded by the source corpus.
Today's systems produce a rapidly exploding amount of data, and the data further derives more data, forming a complex data propagation network that we call the data's lineage. There are many reasons that users want systems to forget certain data including its lineage. From a privacy perspective, users who become concerned with new privacy risks of a system often want the system to forget their data and lineage. From a security perspective, if an attacker pollutes an anomaly detector by injecting manually crafted data into the training data set, the detector must forget the injected data to regain security. From a usability perspective, a user can remove noise and incorrect entries so that a recommendation engine gives useful recommendations. Therefore, we envision forgetting systems, capable of forgetting certain data and their lineages, completely and quickly. This paper focuses on making learning systems forget, the process of which we call machine unlearning, or simply unlearning. We present a general, efficient unlearning approach by transforming learning algorithms used by a system into a summation form. To forget a training data sample, our approach simply updates a small number of summations -- asymptotically faster than retraining from scratch. Our approach is general, because the summation form is from the statistical query learning in which many machine learning algorithms can be implemented. Our approach also applies to all stages of machine learning, including feature selection and modeling. Our evaluation, on four diverse learning systems and real-world workloads, shows that our approach is general, effective, fast, and easy to use.
Subfields: Artificial Intelligence, Privacy & Security, AI Ethics, Fairness & Society
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Percy Liang
- Luke S. Zettlemoyer
- Mohit Bansal
- Quoc V. Le
- Yoshua Bengio
- Jianfeng Gao
- Noah A. Smith
- Hannaneh Hajishirzi
In Privacy & Security
- Abhradeep Thakurta
- Alessandro Achille
- Stefano Soatto
- Aditya Golatkar
- Om Thakkar
- Shuang Song
- Kamalika Chaudhuri
- Nicolas Papernot
In AI Ethics, Fairness & Society
- Aaron Roth
- Alessandro Achille
- Stefano Soatto
- Aditya Golatkar
- Saeed Sharifi-Malvajerdi
- Seth Neel
- Ruby B. Lee
- Zecheng He
Citations in the corpus
- DeepXplore: Automated Whitebox Testing of Deep Learning Systems (2017, background): “Recently, the security and privacy aspects of machine learning have drawn significant attention from the researchers in both machine learning [26, 49, 72] and security [12, 21, 22, 55, 63, 74, 82] communities.”
- Learning with Selective Forgetting (2021, background): “The concept of Machine Unlearning (MU) was first introduced by Cao et al. [Cao and Yang, 2015].”
- DeltaGrad: Rapid retraining of machine learning models (2020, methodology): “Effcient model updating for adding and removing data-points is possible for linear models, based on effcient rank one updates of matrix inverses (e.g., Birattari et al., 1999; Horn & Johnson, 2012; Cao & Yang, 2015, etc).”
- Eternal Sunshine of the Spotless Net: Selective Forgetting in Deep Networks (2019, background): “Machine Unlearning was first studied by [7] in the context of statistical query learning.”
- Making AI Forget You: Data Deletion in Machine Learning (2019, methodology): “They include linear models [55, 27, 83, 81, 18, 74], certain types of lazy learning [88, 6, 11] techniques such as non-parametric Nadaraya-Watson kernel regressions [61] or nearest-neighbors methods [22, 74], recursive support vector machines [19, 81], and co-occurrence based collaborative filtering”
- Measuring Forgetting of Memorized Training Examples (2022, background): “For instance, in machine unlearning [8, 26], a user may issue a “deletion request”, after which their individual contribution to the model must be erased.”
- Certified Data Removal from Machine Learning Models (2019, background): “Removal of specific training samples from models has been studied in prior work on decremental learning (Cauwenberghs and Poggio, 2000; Karasuyama and Takeuchi, 2009; Tsai et al., 2014) and machine unlearning (Cao and Yang, 2015).”
- Unrolling SGD: Understanding Factors Influencing Machine Unlearning (2021, background): “[6] to unlearn datapoints for simple hypothesis spaces which have known SQ learning algorithms.”
- Forgetting Outside the Box: Scrubbing Deep Networks of Information Accessible from Input-Output Observations (2020, background): “onger requirements, enforcing dierential privacy is dicult for deep networks and often results in signicant loss of accuracy [1,9]. Forgetting: The term \machine unlearning" was introduced by [8], who shows an ecient forgetting algorithm in the restricted setting of statistical query learn4 Adi”
- Towards Probabilistic Verification of Machine Unlearning (2020, methodology): “Therefore, Cao and Yang [7] applied ideas from statistical query learning [41] and train the model based on intermediate summations of training data.”
- A Framework for Deprecating Datasets: Standardizing Documentation, Identification, and Communication (2021, background): “For instance, several approaches have been proposed in recent years for ‘machine unlearning’ 1, allowing data to be erased from already trained models [10, 14, 21, 93].”
- DART: Data Addition and Removal Trees (2020, neutral): “More recently, Cao and Yang (2015) first introduced the concept of model unlearning, recognizing that a certain class of models (e.g. naive Bayes) fall under the umbrella of SQ-learning (Kearns 1998), where data deletion is efficient and exact; Schelter (2020) has also developed decremental update…”