Triplet Loss and its Applications to Driver Identification and Verification [WIP]

Abstract

A major issue in driver identification is lack of enough data from the target group of drivers. In this work I use Triplet loss to train neural nets that learn to discriminate between small chunks of driving data. We call this method Deep Driver, through this we significantly reduce the amount of data needed to perform accurate driver identification or verification.

Independent Researcher, CTO

My research interests include time series classification and privacy-preserving machine learning.