Beyond Differential Privacy 2019
2019-05-06..2019-05-10
Berkeley, United States
as the organisers put it
The notion of differential privacy has provided a firm foundation for the formal study of privacy guarantees. Building on this foundation, this workshop will examine ideas in the space of data rights, including fairness, accountability, and privacy (beyond differential privacy), with a goal of surfacing problems that would benefit from the attention of the CS theory lens. The workshop brings together experts from theoretical computer science, law, statistics, economics, and the social sciences more broadly.
who it's for
privacy researchers, machine learning researchers, algorithmic fairness researchers, legal scholars and regulators, data scientists working with sensitive data, technology policy makers
topics
Gaussian differential privacy; data deletion for trained models; metric learning for individual fairness; preference-informed fairness formulations; information-content audits for fairness; privacy dependencies and population inference; protective optimization technologies; shuffle model and federated aggregation; hardware enclaves for privacy; Bayesian privacy-loss mechanisms (KL/mutual information); accountability and contestable algorithms; GDPR and CCPA impacts on data markets
| title | author | length |
|---|---|---|
| Societal Concerns in Targeted Advertising | Aleksandra Korolova | 01h 05m |
Three concrete failures of targeted ads and steps to fix them: custom audiences and Audience Insights let an uploader infer a single person’s attributes… | ||
| Getting AI to Forget About You: Data Deletion From Machine Learning Models | James Zou | 31m |
Shows that trained models must sometimes remove individual records on demand, and formalizes deletion as a mapping that returns the model as if the point were… | ||
| Optimal Privacy-Constrained Mechanisms | Xiaosheng Mu | 34m |
Work introduces a Bayesian privacy-loss metric measured by KL divergence (equivalently mutual information) and enforces an exogenous privacy budget kappa on… | ||
| Checks and Balances in Governance by AI | Niva Elkin-Koren | 01h 06m |
Argument: liberal democracies designed for rule-based government are failing when firms govern by data and algorithms. Uses Waze routing to show externalities… | ||
| Algorithm Design and Law: Potential and Challenges | Inbal Talgam-Cohen | 01h 11m |
Asks whether algorithm design can encode legal doctrines and gives concrete examples: a formal model called Responsibility-Sensitive Safety (RSS) translates… | ||
| Preference Informed Fairness | Gal Yona | 31m |
Proposes preference-informed individual fairness: classifiers may deviate from individual-fair outcomes only when each affected person prefers the deviation to… | ||
| Tracking and Improving Information in the Service of Fairness | Michael Kim | 38m |
Paper shows that disparities often come from unequal information content in calibrated risk scores, not from an unavoidable tradeoff between accuracy and… | ||
| Privacy Dependencies | Solon Barocas | 01h 04m |
Argues that privacy about one person often depends on others, and gives a three‑part taxonomy: tie‑based (familial searches like Golden State Killer via… | ||
| Unwanted Learning: What It Is, Why It Matters, and What To Do About It | Vitaly Shmatikov | 01h 08m |
Empirical results show that simple classifiers trained on one label routinely learns identities from gender models and extract many unlabeled attributes (race… | ||
| Beyond Software: Hardware-Based Security | Ilya Mironov | 32m |
Argues that replacing a trusted curator with hardware and protocol layers yields practical privacy: the shuffle model amplifies local DP into stronger central… | ||
| Gaussian Differential Privacy (GDP) | Jinshuo Dong | 31m |
Proposes Gaussian Differential Privacy as a replacement for single (ε,δ) pairs, defining privacy by a trade-off function between type I and II errors and then… | ||
| Protective Optimization Technologies | Seda Guerses | 01h 04m |
Argument that optimization systems (cloud, feedback loops, objective functions) create externalities that privacy fixes like differential privacy don’t stop… | ||
| Metric Learning for Individual Fairness | Christina Ilvento | 43m |
Proposes learning a distance metric from human judgments by querying a small set of representative points, using cheap triplet (is A closer to R or B?)… | ||
| Making Data Great Again | Julia Lane | 01h 02m |
Describes building an Integrated Data Infrastructure (IDI) in New Zealand and a US Administrative Data Research Facility that links health, education, justice… | ||
| Contextual Online False Discovery Rate Control | Shiva Kasiviswanathan | 38m |
Introduces a formal model of contextual online false discovery rate control, defines contextual generalized alpha-investing procedures and a weighted LORD… | ||
| Viability of a Privacy-differentiated Market for Free Online Services | Lalitha Sankar | 39m |
Models two free-service rivals (Google vs DuckDuckGo) as a Hotelling duopoly model on an epsilon privacy line with linear cost/revenue, uniform consumer types… | ||
| Strategic Goals and Mechanisms of the GDPR and CCPA | Chris Hoofnagle | 01h 07m |
Comparative analysis of GDPR and CCPA that pinpoints concrete rules and enforcement outcomes: CCPA allows financial inducements with limits and forces firms to… | ||
| Assumptions of Computer Science Interacting with Assumptions of Text-driven Normativities | Mireille Hildebrandt | 01h 09m |
Argues that privacy as protection of the incomputable self means formal tools like differential privacy and homomorphic encryption only solve reidentification… | ||
| On Valuing and Procuring Private Data | Bo Waggoner | 33m |
Argues personal data is best framed as controllable rights (like copyright) and proposes an online procurement mechanism that posts take-it-or-leave-it price… | ||
| Individual Statistical Fairness | Aaron Roth | 30m |
Shows that minimizing average classification error can systematically exclude a minority group even when that group is more qualified, and proposes a formal… | ||
| Accountability in Computing | Rebecca Wright | 57m |
Defines accountability as a trace-driven utility definition and places solutions on a prevention-to-punishment spectrum, formalizing systems as traces plus a… | ||
| nothing matches | ||