1. Regret Minimizing Audits: A Learning-theoretic Basis for Privacy Protection Jeremiah Blocki, Nicolas Christin, Anupam Datta, Arunesh Sinha Carnegie Mellon University

2. Motivation Goal: treatment Rigid access control hinders treatment Permissive access control ⇒ human in the loop ⇒ privacy violations Breach 2

3. A real problem 3

4. Enforcement Using Audits Permissive access control ◦ If in doubt allow access Log the accesses Review the accesses later and find violations Adhoc approaches in practice ◦ FairWarning audit tool implements simple heuristics, e.g., flag all celebrity access 4

5. Humans are the weakest link 5 Design an audit mechanism robust to irresponsible/careless human behavior Audit based enforcement involves human participation ◦ Employees trusted to act responsibly

6. Desiderata Principled study of the audit process ◦ A model (including humans) for audit process ◦ Properties of the audit mechanism ◦ Audit mechanism which provably satisfies the property 6

7. Audit Algorithm by Example OverviewAudit ModelLow Regret Algorithm Auditing budget: $3000/ cycle Cost for one inspection: $ 1 00 Only 30 inspections per cycle Auditor 1 00 accesses 30 accesses 70 accesses Access divided into 2 types Reputation Loss from 1 violation (internal, external) $500, $ 1 000 $250, $500

8. Audit Algorithm Choices 8 Only 30 inspections Employee incentive unknown 0 10102030 20 10100 Consider 4 possible allocations of the available 30 inspections 1.0 Weights Choose allocation probabilistically based on weights OverviewAudit ModelLow Regret Algorithm

9. No. of Access Audit Algorithm Run 9 0 10102030 20 10100 0.5 2.01.5 Updated weights Observed Loss $2000$ 1 500$ 1 000 $750$ 1 250 $ 1 500 Learn from experience: weights updated using observed and estimated loss 2 4 Actual Violation Ext. Caught Int. Caught 1 1 1 2 30 70 OverviewAudit ModelLow Regret Algorithm Estimated Loss

10. Main Contributions  A game model for the audit process  Defining a desirable property of audit mechanisms, namely low regret  An efficient audit mechanism RMA that provably achieves low regret o Better bound on regret than existing algorithms that achieve low regret 10 OverviewAudit ModelLow Regret Algorithm

11. Repeated Game Model Game model Typical actions in one round ◦ Emp action: (access, violate) = ([30,70], [2,4]) ◦ Org action: inspection = ([ 1 0,20]) Inspect Reputation loss Audit Cost Access, Violate One audit cycle (round) 11 Imperfection OverviewAudit ModelLow Regret Algorithm

12. Game Payoffs Organization’s payoff ◦ Audit cost depends on the number of inspections ◦ Reputation loss depends on the number of violations caught Employee’s payoff unknown ◦ Guarantees of any audit mechanism in this model holds irrespective of behavior of the employee Reputation loss Audit cost 12 OverviewAudit ModelLow Regret Algorithm

13. Regret by Example $5 $6 $0$5 1 2 3,1 3, 2 Payoff of Org only Players Emp Org: s Round 1 3, 1 2 ( 1 ) Round 2 3,2 1 (-5) Total Payoff Unknown -4 Org : s 1 1 (2) 1 (-5) -3 Total regret(s, s 1 ) = (–5) – (–6) = 1 regret(s, s 1 ) = 1 /2 Strategy: outputs an action for every round Emp Org 13 Players Emp Org:s Round 1 3, 1 2 ( $6 ) Round 2 3, 2 1 ( $0) Total Payoff Unknown $6 Org:s 1 1 ($5) 1( $0) $5 OverviewAudit ModelLow Regret Algorithm

14. Meaning of Regret Low regret of s w.r.t. s 1 means s performs as well as s 1 Desirable property of an audit mechanism ◦ Low regret w.r.t all strategies in a given set of strategies ◦ regret → 0 as T → ∞ 14 OverviewAudit ModelLow Regret Algorithm

15. Regret minimization Multiplicative weight update (MWU) ◦ standard algorithm that achieves low regret w.r.t. to all strategies in a given set The regret bound of MWU is  N: number of strategies in the given set  T: number of rounds of the game  All payoffs scaled to lie in [0, 1 ] Why not MWU? ◦ Imperfect information, unavailable strategies 15 OverviewAudit ModelLow Regret Algorithm

16. Regret Minimizing Audits (RMA) 16 New audit cycle starts. Find AWAKE Pick s in AWAKE with probability D t (s) ∝ w s Update weight* of strategies s in AWAKE Estimate payoff vector Pay using Pay(s) Violation caught; obtain payoff Pay(s) w s = 1 for all strategies s * OverviewAudit ModelLow Regret Algorithm

17. Guarantees of RMA With probability RMA achieves the regret bound ◦ N is the set of strategies ◦ T is the number of rounds ◦ All payoffs scaled to lie in [0, 1 ] Better bound than any existing algorithm 17 OverviewAudit ModelLow Regret Algorithm

18. Take Away Message Future Work ◦ Evaluation over real hospital audit logs ◦ Analyze performance with more complex adversary models  Worst case + rational Optimize costs, given imperfect nature of periodic audits in a setting with adaptive adversaries (employees) whose incentives are not known. 18

19. 19