Challenges in Scaling Privacy Enhancing Technologies to Campus Level IoT Testbed
Talkat Data Privacy: From Foundations to Applications (Simons Institute workshop, 2019) · 2019-03-04..03-08·
summary
Campus IoT testbed TIPPERS runs in 35 buildings and ingests roughly ~1M location events/day (≈44k/day per building), integrates Pegasus for differential privacy and Pulsar/Jana (Galois) for encrypted storage, and connects to Honeywell and SPAWAR deployments. Differentially private occupancy with epsilon=1 still lets models localize about 4–5 of the top 50 visitors (vs ~2–3 from priors). Pure MPC is infeasible at scale (MPC ~10^6× slower); query-binning that separates sensitive/non-sensitive data yields large speedups while reducing leakage.
AI-written summary · about
chapters (17)
- 0:00Intro
- 1:28Internet of Things: the next big wave Systems that empower everyday physical devices to connect to the internet and to
- 2:52TIPPERS - Testbed Concept
- 4:20Talk Outline
- 6:10TIPPERS Deployments
- 8:09UCI Testbed: At Scale WiFi Connectivity-based Testbed
- 10:42Applications
- 12:45Secure Data Management via Mediation-based Architecture
- 13:50Inferences from Anonymized Mac Addresses of Devices Connected to WiFI AP
- 19:38Experimental Setup
- 22:49Aggregate Privacy Metric
- 24:32Challenges @ Campus Scale
- 27:09Query Execution Cost on Outsourced Data
- 28:43Key Insight ...partial sensitivity of data
- 35:40Query Binning Extensions
- 37:40One Sided Differential Privacy - OSDP
- 38:27Interesting Side Effect of OSDP...
related talks
31mScaling privacy embedded distributed ML systems
01h 19mWhen Data Can’t Move but Insight Must
46mPrivacy Preserving IoT
16mBalancing Privacy: Leveraging Privacy Budgets for Privacy-Enhancing Technologies
01h 16mCommercial opportunities for designing privacy in: Embedding privacy ...
27mMetadata-private data transfer
26mDoes using data mean giving up privacy?
58mLarge-Scale Private Learning, Part III