Lecture 13C: Differentially Private Machine Learning - Gradient Perturbation

preview_player
Показать описание
  1. Gautam Kamath
  2. differential privacy
  3. lecture
  4. class
  5. course
Рекомендации по теме
Lecture 13C: Differentially 0:37:49

Lecture 13C: Differentially Private Machine Learning - Gradient Perturbation

Learning Differentially Private 0:01:15

Learning Differentially Private Mechanisms

Lecture 13B: Differentially 0:24:19

Lecture 13B: Differentially Private Machine Learning - Output and Objective Perturbation

Differentially Private Data 0:37:06

Differentially Private Data Generative Models and Safety-Critical Scenario Generation for... | Bo Li

USENIX Security '23 0:13:13

USENIX Security '23 - Tight Auditing of Differentially Private Machine Learning

Antti Honkela: Accurate 0:44:15

Antti Honkela: Accurate privacy accounting for differentially private machine learning

Lecture 13A: Differentially 0:43:54

Lecture 13A: Differentially Private Machine Learning - A Quick Primer

Differentially Private Top-k 0:07:41

Differentially Private Top-k Selection via Stability on Unknown Domain

[1A] Differentially Private 0:15:38

[1A] Differentially Private Naive Bayes Classifier using Smooth Sensitivity

Lecture 13: The 0:32:50

Lecture 13: The Matrix Mechanism (Part 1)

Towards Practical Differentially 0:01:01

Towards Practical Differentially Private Convex Optimization

[5C] Differentially Private 0:13:58

[5C] Differentially Private Speaker Anonymization

Differentially Private Histograms 0:00:59

Differentially Private Histograms in the Shuffle Model from Fake Users

ARES 2021 - 0:13:54

ARES 2021 - Efficient Error Prediction for Differentially Private Algorithms

Differentially private Bayesian 0:03:15

Differentially private Bayesian learning on distributed data, NIPS 2017

Differentially Private Histograms 0:16:45

Differentially Private Histograms in the Shuffle Model from Fake Users

Improving the Privacy 0:38:18

Improving the Privacy Utility Tradeoff in Differentially Private Machine Learning with Public Data

PEPR '24 - 0:13:12

PEPR '24 - Utility Analysis for Differentially-Private Pipelines

AdKDD 2024: Training 0:10:00

AdKDD 2024: Training Differentially Private Ad Prediction Models With Semi-Sensitive Features

Vladimir Dvorkin: Differentially 0:13:32

Vladimir Dvorkin: Differentially Private Distributed Optimal Power Flow

[1A] DPlis: Boosting 0:17:42

[1A] DPlis: Boosting Utility of Differentially Private Deep Learning via Randomized Smoothing

AISTATS 2012: A 0:20:18

AISTATS 2012: A Differentially Private Stochastic Gradient Descent Algorithm for Multiparty...

How to Create 0:13:21

How to Create Differentially Private Synthetic Data

Joshua Falk: Generating 0:36:13

Joshua Falk: Generating realistic, differentially private data sets using GANs | PyData NYC 2019

Комментарии
Автор

I think C needs to be a convex set, no?

screwnacorn
Автор

26:02 Why log(n/delta) is thrown away here?

wap
Автор

Thanks a lot for making this course available online. I have a question in @37.40, why delta is not multiplied by n (which comes from [ 1/n from sampling * T =n^2 from the basic composition). The same for strong composition thm.

ahmedroushdy
Автор

towards @37, I think it should be T =~ n^2, not Tn =~ n^2

z_doc