O SlideShare utiliza cookies para otimizar a funcionalidade e o desempenho do site, assim como para apresentar publicidade mais relevante aos nossos usuários. Se você continuar a navegar o site, você aceita o uso de cookies. Leia nosso Contrato do Usuário e nossa Política de Privacidade.

O SlideShare utiliza cookies para otimizar a funcionalidade e o desempenho do site, assim como para apresentar publicidade mais relevante aos nossos usuários. Se você continuar a utilizar o site, você aceita o uso de cookies. Leia nossa Política de Privacidade e nosso Contrato do Usuário para obter mais detalhes.

O slideshow foi denunciado.

Gostou da apresentação? Compartilhe-a!

- Scott Clark, CEO, SigOpt, at The AI... by MLconf 572 views
- Nikhil Garg, Engineering Manager, Q... by MLconf 1264 views
- Josh Wills, Head of Data Engineerin... by MLconf 1107 views
- Andrew Musselman, Committer and PMC... by MLconf 906 views
- Scott Clark, Co-Founder and CEO, Si... by MLconf 1017 views
- Sanjeev Satheesj, Research Scientis... by MLconf 398 views

551 visualizações

Publicada em

Abstract Summary:

Differential Privacy and Machine Learning:

In this talk, we will give a friendly introduction to Differential Privacy, a rigorous methodology for analyzing data subject to provable privacy guarantees, that has recently been widely deployed in several settings. The talk will specifically focus on the relationship between differential privacy and machine learning, which is surprisingly rich. This includes both the ability to do machine learning subject to differential privacy, and tools arising from differential privacy that can be used to make learning more reliable and robust (even when privacy is not a concern).

Publicada em:
Tecnologia

Sem downloads

Visualizações totais

551

No SlideShare

0

A partir de incorporações

0

Número de incorporações

6

Compartilhamentos

0

Downloads

19

Comentários

0

Gostaram

2

Nenhuma incorporação

Nenhuma nota no slide

- 1. Differential Privacy And Machine Learning Aaron Roth March 24, 2017
- 2. Protecting Privacy is Important Class action lawsuit accuses AOL of violating the Electronic Communications Privacy Act, seeks $5,000 in damages per user. AOL’s director of research is fired.
- 3. Protecting Privacy is Important Class action lawsuit (Doe v. Netflix) accuses Netflix of violating the Video Privacy Protection Act, seeks $2,000 in compensation for each of Netflix’s 2,000,000 subscribers. Settled for undisclosed sum, 2nd Netflix Challenge is cancelled.
- 4. Protecting Privacy is Important The National Human Genome Research Institute (NHGRI) immediately restricted pooled genomic data that had previously been publically available.
- 5. So What is Differential Privacy?
- 6. So What is Differential Privacy? • Differential Privacy is about promising people freedom from harm. “An analysis of a dataset D is differentially private if the data analyst knows almost no more about Alice after the analysis than he would have known had he conducted the same analysis on an identical data set with Alice’s data removed.”
- 7. D Differential Privacy [Dwork-McSherry-Nissim-Smith 06] Algorithm Pr [r] ratio bounded Alice Bob Chris Donna ErnieXavier
- 8. Differential Privacy 𝑋: The data universe. 𝐷 ⊂ 𝑋: The dataset (one element per person) Definition: Two datasets 𝐷, 𝐷′ ⊂ 𝑋 are neighbors if they differ in the data of a single individual.
- 9. Differential Privacy 𝑋: The data universe. 𝐷 ⊂ 𝑋: The dataset (one element per person) Definition: An algorithm 𝑀 is 𝜖-differentially private if for all pairs of neighboring datasets 𝐷, 𝐷′ , and for all outputs x: Pr 𝑀 𝐷 = 𝑥 ≤ (1 + 𝜖) Pr 𝑀 𝐷′ = 𝑥
- 10. Some Useful Properties Theorem (Postprocessing): If 𝑀(𝐷) is 𝜖-private, and 𝑓 is any (randomized) function, then 𝑓(𝑀 𝐷 ) is 𝜖-private.
- 11. So… 𝑥 = Definition: An algorithm 𝑀 is 𝜖-differentially private if for all pairs of neighboring datasets 𝐷, 𝐷′ , and for all outputs x: Pr 𝑀 𝐷 = 𝑥 ≤ (1 + 𝜖) Pr 𝑀 𝐷′ = 𝑥
- 12. So… 𝑥 = Definition: An algorithm 𝑀 is 𝜖-differentially private if for all pairs of neighboring datasets 𝐷, 𝐷′ , and for all outputs x: Pr 𝑀 𝐷 = 𝑥 ≤ (1 + 𝜖) Pr 𝑀 𝐷′ = 𝑥
- 13. So… 𝑥 = Definition: An algorithm 𝑀 is 𝜖-differentially private if for all pairs of neighboring datasets 𝐷, 𝐷′ , and for all outputs x: Pr 𝑀 𝐷 = 𝑥 ≤ (1 + 𝜖) Pr 𝑀 𝐷′ = 𝑥
- 14. Some Useful Properties Theorem (Composition): If 𝑀1, … , 𝑀 𝑘 are 𝜖-private, then: 𝑀 𝐷 ≡ (𝑀1 𝐷 , … , 𝑀 𝑘 𝐷 ) is 𝑘𝜖-private.
- 15. So… You can go about designing algorithms as you normally would. Just access the data using differentially private “subroutines”, and keep track of your “privacy budget” as a resource. Private algorithm design, like regular algorithm design, can be modular.
- 16. Some simple operations: Answering Numeric Queries Def: A numeric function 𝑓 has sensitivity 𝑐 if for all neighboring 𝐷, 𝐷′ : 𝑓 𝐷 − 𝑓 𝐷′ ≤ 𝑐 Write 𝑠 𝑓 ≡ 𝑐 • e.g. “How many software engineers are in the room?” has sensitivity 1. • “What fraction of people in the room are software engineers?” has sensitivity 1 𝑛 .
- 17. Some simple operations: Answering Numeric Queries The Laplace Mechanism: 𝑀𝐿𝑎𝑝 𝐷, 𝑓, 𝜖 = 𝑓 𝐷 + 𝐿𝑎𝑝 𝑠 𝑓 𝜖 Theorem: 𝑀𝐿𝑎𝑝(⋅, 𝑓, 𝜖) is 𝜖-private.
- 18. Some simple operations: Answering Numeric Queries The Laplace Mechanism: 𝑀𝐿𝑎𝑝 𝐷, 𝑓, 𝜖 = 𝑓 𝐷 + 𝐿𝑎𝑝 𝑠 𝑓 𝜖 Theorem: The expected error is 𝑠 𝑓 𝜖 (can answer “what fraction of people in the room are software engineers?” with error 0.2%)
- 19. Some simple operations: Answering Non-numeric Queries “What is the modal eye color in the room?” 𝑅 = {Blue, Green, Brown, Red} • If you can define a function that determines how “good” each outcome is for a fixed input: – E.g. 𝑞 𝐷, Red =“fraction of people in D with red eyes”
- 20. Some simple operations: Answering Non-numeric Queries 𝑀 𝐸𝑥𝑝(𝐷, 𝑅, 𝑞, 𝜖): Output 𝑟 ∈ 𝑅 w.p. ∝ 𝑒2𝜖⋅𝑞 𝐷,𝑟 Theorem: 𝑀 𝐸𝑥𝑝(𝐷, 𝑅, 𝑞, 𝜖) is 𝑠 𝑞 ⋅ 𝜖-private, and outputs 𝑟 ∈ 𝑅 such that: 𝐸 𝑞 𝐷, 𝑟 − max 𝑟∗∈𝑅 𝑞 𝐷, 𝑟∗ ≤ 2𝑠 𝑞 𝜖 ⋅ ln 𝑅 (can find a color that has frequency within 0.5% of the modal color in the room)
- 21. So what can we do with that? Empirical Risk Minimization: *i.e. almost all of supervised learning Find 𝜃 to minimize: 𝐿 𝜃 = 𝑖=1 𝑛 ℓ 𝜃, 𝑥𝑖, 𝑦𝑖
- 22. So what can we do with that? Empirical Risk Minimization: Simple Special Case: Linear Regression Find 𝜃 ∈ 𝑅 𝑑 to minimize: 𝑖=1 𝑛 𝜃, 𝑥𝑖 − 𝑦𝑖 2
- 23. So what can we do with that? • First Attempt: Try the exponential mechanism! • Define 𝑞 𝜃, 𝐷 = − 𝑋𝜃 − 𝑦 𝑇 (𝑋𝜃 − 𝑦) • Output 𝜃 ∈ ℝ 𝑑 w.p. ∝ 𝑒2𝜖𝑞(𝜃,𝐷) – How? • In this case, reduces to: Output 𝜃∗ + 𝑁 0, 𝑋 𝑇 𝑋 −1 2𝜖 Where 𝜃∗ = arg min 𝑖=1 𝑛 𝜃, 𝑥𝑖 − 𝑦𝑖 2
- 24. So what can we do with that? Empirical Risk Minimization: The General Case (e.g. deep learning) Find 𝜃 to minimize: 𝐿 𝜃 = 𝑖=1 𝑛 ℓ 𝜃, 𝑥𝑖, 𝑦𝑖
- 25. Stochastic Gradient Descent Convergence depends on the fact that at each round: 𝔼 𝑔𝑡 = 𝛻L(𝜃) Let 𝜃1 = 0 𝑑 For 𝑡 = 1 to 𝑇: Pick 𝑖 at random. Let 𝑔𝑡 ← 𝛻ℓ 𝜃 𝑡, 𝑥𝑖, 𝑦𝑖 Let 𝜃 𝑡+1 ← 𝜃 𝑡 − 𝜂 ⋅ 𝑔𝑡
- 26. Private Stochastic Gradient Descent Still have: 𝔼 𝑔𝑡 = 𝛻L 𝜃 ! (Can still prove convergence theorems, and run the algorithm…) Privacy guarantees can be computed from: 1) The privacy of the Laplace mechanism 2) Preservation of privacy under post-processing, and 3) Composition of privacy guarantees. Let 𝜃1 = 0 𝑑 For 𝑡 = 1 to 𝑇: Pick 𝑖 at random. Let 𝑔𝑡 ← 𝛻ℓ 𝜃 𝑡, 𝑥𝑖, 𝑦𝑖 + 𝐿𝑎𝑝 𝜎 𝑑 Let 𝜃 𝑡+1 ← 𝜃 𝑡 − 𝜂 ⋅ 𝑔𝑡
- 27. What else can we do? • Statistical Estimation • Graph Analysis • Combinatorial Optimization • Spectral Analysis of Matrices • Anomaly Detection/Analysis of Data Streams • Convex Optimization • Equilibrium computation • Computation of optimal 1-sided and 2-sided matchings • Pareto Optimal Exchanges • …
- 28. Differential Privacy ⇒ Learning Theorem*: An 𝜖-differentially private algorithm cannot overfit its training set by more than 𝜖. *Lots of interesting details missing! See our paper in Science.
- 29. from numpy import * def Thresholdout(sample, holdout, q, sigma, threshold): sample_mean = mean([q(x) for x in sample]) holdout_mean = mean([q(x) for x in holdout]) if (abs(sample_mean - holdout_mean) < random.normal(threshold, sigma)): # q does not overfit return sample_mean else: # q overfits return holdout_mean + random.normal(0, sigma) thresholdout.py: Thresholdout [DFHPRR15]
- 30. Reusable holdout example • Data set with 2n = 20,000 rows and d = 10,000 variables. Class labels in {-1,1} • Analyst performs stepwise variable selection: 1. Split data into training/holdout of size n 2. Select “best” k variables on training data 3. Only use variables also good on holdout 4. Build linear predictor out of k variables 5. Find best k = 10,20,30,…
- 31. Classification after feature selection No signal: data are random gaussians labels are drawn independently at random from {-1,1} Thresholdout correctly detects overfitting!
- 32. Strong signal: 20 features are mildly correlated with target remaining attributes are uncorrelated Thresholdout correctly detects right model size! Classification after feature selection
- 33. So… • Differential privacy provides: – A rigorous, provable guarantee with a strong privacy semantics. – A set of tools and composition theorems that allow for modular, easy design of privacy preserving algorithms. – Protection against overfitting even when privacy is not a concern.
- 34. Thanks! To learn more: • Our textbook on differential privacy: – Available for free on my website: http://www.cis.upenn.edu/~aaroth • Connections between Privacy and Overfitting: – Dwork, Feldman, Hardt, Pitassi, Reingold, Roth, “The Reusable Holdout: Preserving Validity in Adaptive Data Analysis”, Science, August 7 2015. – Dwork, Feldman, Hardt, Pitassi, Reingold, Roth, “Preserving Statistical Validity in Adaptive Data Analysis”, STOC 2015. – Bassily, Nissim, Stemmer, Smith, Steinke, Ullman, “Algorithmic Stability for Adaptive Data Analysis”, STOC 2016. – Rogers, Roth, Smith, Thakkar, “Max Information, Differential Privacy, and Post-Selection Hypothesis Testing”, FOCS 2016. – Cummings, Ligett, Nissim, Roth, Wu, “Adaptive Learning with Robust Generalization Guarantees”, COLT 2016.

Nenhum painel de recortes público que contém este slide

Parece que você já adicionou este slide ao painel

Criar painel de recortes

Seja o primeiro a comentar