Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FedDriveScore: Federated Scoring Driving Behavior with a Mixture of Metric Distributions

Lin Lu

TL;DR

FedDriveScore presents an unsupervised, privacy-preserving approach to scoring driving behavior by modeling each metric as a distribution and aggregating via a CRITIC-weighted mixture (CRITIC-DM). It implements a consistently federated version using homomorphic encryption and secure histograms to mitigate non-IID data and privacy issues, aiming to match centralized learning performance. The framework is validated on real fleet data and virtual UBI data, showing convergence of global statistics and weights and high utility consistency with CL, plus objective fairness relative to subjective scores. The work contributes a practical, scalable solution for driver profiling in IoV contexts with strong privacy guarantees.

Abstract

Scoring the driving performance of various drivers on a unified scale, based on how safe or economical they drive on their daily trips, is essential for the driver profile task. Connected vehicles provide the opportunity to collect real-world driving data, which is advantageous for constructing scoring models. However, the lack of pre-labeled scores impede the use of supervised regression models and the data privacy issues hinder the way of traditionally data-centralized learning on the cloud side for model training. To address them, an unsupervised scoring method is presented without the need for labels while still preserving fairness and objectiveness compared to subjective scoring strategies. Subsequently, a federated learning framework based on vehicle-cloud collaboration is proposed as a privacy-friendly alternative to centralized learning. This framework includes a consistently federated version of the scoring method to reduce the performance degradation of the global scoring model caused by the statistical heterogeneous challenge of local data. Theoretical and experimental analysis demonstrate that our federated scoring model is consistent with the utility of the centrally learned counterpart and is effective in evaluating driving performance.

FedDriveScore: Federated Scoring Driving Behavior with a Mixture of Metric Distributions

TL;DR

FedDriveScore presents an unsupervised, privacy-preserving approach to scoring driving behavior by modeling each metric as a distribution and aggregating via a CRITIC-weighted mixture (CRITIC-DM). It implements a consistently federated version using homomorphic encryption and secure histograms to mitigate non-IID data and privacy issues, aiming to match centralized learning performance. The framework is validated on real fleet data and virtual UBI data, showing convergence of global statistics and weights and high utility consistency with CL, plus objective fairness relative to subjective scores. The work contributes a practical, scalable solution for driver profiling in IoV contexts with strong privacy guarantees.

Abstract

Scoring the driving performance of various drivers on a unified scale, based on how safe or economical they drive on their daily trips, is essential for the driver profile task. Connected vehicles provide the opportunity to collect real-world driving data, which is advantageous for constructing scoring models. However, the lack of pre-labeled scores impede the use of supervised regression models and the data privacy issues hinder the way of traditionally data-centralized learning on the cloud side for model training. To address them, an unsupervised scoring method is presented without the need for labels while still preserving fairness and objectiveness compared to subjective scoring strategies. Subsequently, a federated learning framework based on vehicle-cloud collaboration is proposed as a privacy-friendly alternative to centralized learning. This framework includes a consistently federated version of the scoring method to reduce the performance degradation of the global scoring model caused by the statistical heterogeneous challenge of local data. Theoretical and experimental analysis demonstrate that our federated scoring model is consistent with the utility of the centrally learned counterpart and is effective in evaluating driving performance.
Paper Structure (27 sections, 3 theorems, 17 equations, 13 figures, 12 tables, 1 algorithm)

This paper contains 27 sections, 3 theorems, 17 equations, 13 figures, 12 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Unsupervised Scoring Methods
  4. Federated Learning
  5. Knowledge Gap
  6. Problem Formulation
  7. Proposed Solution
  8. FedDriveScore Framework
  9. CF4CRITIC-DM method
  10. Definition of Evaluation Metrics
  11. Observation of Metric Distributions
  12. Model Training Process
  13. Model inferring Process
  14. Convergence Analysis
  15. Privacy Analysis
  16. ...and 12 more sections

Key Result

Theorem 4.1

Assume $X$ a sample population in a homogeneous space $\mathbb{R}^d$, and its population mean and variance are $\mathbb{E}[X]$ and $\mathbb{D}[X]$ respectively . For each iteration $t$, it can be considered that $\mathcal{B}^t$ is the result of a random sampling of $X$, and $\boldsymbol{\mu}^t$ and where $\boldsymbol{\mu}^t$ and $\mathbb{E}[X]$ are the estimated and real mean vector of $X$, respe

Figures (13)

  • Figure 1: The overview of FedDriveScore Framework via IoV.
  • Figure 2: Illustration of secure histogram for observing metric distributions.
  • Figure 3: The overall distributions of metrics in the fleet driving data.
  • Figure 4: A client-level histograms of metrics in the fleet driving data.
  • Figure 5: The estimated mean and variance of the metrics in the fleet driving data.
  • ...and 8 more figures

Theorems & Definitions (8)

  • Definition 2.1
  • Definition 2.2: $\delta$-Accuracy
  • Theorem 4.1
  • Proof 4.1
  • Theorem 4.2
  • Proof 4.2
  • Definition 4.1: Semi-honest Adversary
  • Theorem 4.3