PRISMA builds differentially private synthetic in-car energy traces
Original title: PRISMA: Differentially Private Synthetic In-Car Telemetry via Differentiable Simulation
At epsilon 0.5, PRISMA beats tabular DP baselines by keeping diverse power profiles instead of collapsing to few modes.
PRISMA encodes in-car energy traces into a low-dimensional latent space, synthesizes that distribution with the differentially private tabular algorithm AIM, and decodes samples through a differentiable physics simulation. Decoding uses public map data, elevation, and a vehicle energy model, so it spends no extra privacy budget. The authors report better synthesis performance than prior tabular and marginal-based DP baselines at epsilon 0.5.
- Context variables include start point, end point, car model, and driving behavior.
- Latent vectors are synthesized with the differentially private tabular algorithm AIM.
- Decoding uses public maps and a physics model without extra privacy budget.
- Authors report preserved profile diversity rather than collapse to a few modes.
From the original article
1 min read at arxiv.orgDifferentially private synthesis of in-car energy consumption data is challenging because the output is high-dimensional, and known lower bounds show that the privacy-utility trade-off degrades with the square root of the number of output dimensions.
By Kilian Zeiseweis · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.