ConceptAI & ML
Differential privacy
A mathematical definition of privacy ensuring that an algorithm's output does not reveal whether any single individual's data was included in the dataset.
Know first
Named as a prerequisite by the lectures that define this. The number is how many say so. A grey one has no page yet.
Probability distribution6Randomized algorithm4Datasets2Probability2Auxiliary information1Data privacy1Indistinguishability1Randomized mechanisms1
Opens up
Concepts that name this one as a prerequisite.
Lecturers
9Each in their own terms, from the easiest level.
Michael I. Jordan
A privacy method that adds noise to data while preserving useful aggregate patterns.Cynthia Dwork
A definition stating that the outcome of any statistical analysis is essentially equally likely whether any individual joins or refrains from joining the dataset.Charlie Hou
A mathematical framework ensuring that the output of an algorithm does not change significantly if any single individual's data is added to or removed from the training dataset.Adam Dziedzic
Produces roughly same or indistinguishable outputs on any pair of prompt datasets that differ only by a single data point.Connor Wagaman
A definition imposing a restriction on algorithms where running that algorithm on data sets that differ by one individual results in outputs that are hard to tell apart.Andy Dong
A mathematical framework for measuring the privacy guarantees of algorithms, defined here via hockey-stick divergence over adjacent datasets.Kai Chen
An algorithm M satisfies epsilon-delta DP if neighboring datasets differing in one record produce very close output distributions controlled by epsilon and delta.Katrina Ligett
A mathematical framework for measuring the privacy loss of a randomized algorithm when individual data points are changed.
Lectures that assume it
3Connections
20The lectures that define it or assume it, and the concepts next to it.
Concepts8
- Probability distribution
- Randomized algorithm
- Probability
- Gaussian mechanism
- Privacy auditing
- $(\alpha, \beta)$-accuracy
Show 2 moreShow fewer
Lectures12
- Local Node Differential Privacy
- A Collectivist, Economic Perspective on AI Systems
- Differential Privacy: Fundamentals to Forefront I
- POPRi: Private Federated Learning using Preference-Optimized Synthetic Data
- Privacy Amplification for Correlated-Noise Mechanisms via b-Min-Sep Subsampling
- Private Adaptations of Large Language Models
Show 6 moreShow fewer
- Differentially Private Table-Image Multimodal Data Generation
- Stable Estimators for Fast Private Statistics
- Foundations, Privacy & Generalization Challenges in Modern Machine Learning
- Mechanism Design & Game-Theoretic Incentives for Collaborative Learning
- Leveraging Per-Instance Privacy for Machine Unlearning
- Worst-Case Membership Inference of Language Models