Dimitra Politaki – Neo
Some intuition about l2 (“Ridge”) and l1 (“Lasso”) regularization
Kostia Maksymenko – Athena
One finds models with l2 and l1 regularization in statistics, machine learning, signal processing and other domains of applied mathematics. This talk will be interesting for those of you who use it like a “black box”, i.e. without understanding the principals and hypothesis lying behind them. The objective of this presentation is to give some principal properties of these problems in a relatively simple language. I will talk about probabilistic and geometrical interpretation of l2 and l1 regularization which will allow you to better understand when and why does it work (or not).
Impact of GPS-enabled routing applications on traffic
Nicolas Laurent-Brouty – Acumes
My presentation will detail the impact of the increasing penetration of routing apps on road usage. To model the issue, we distinguish two distinct classes of drivers. The first category has full knowledge of the network and the traffic conditions, while the other only has limited knowledge. The framework applies both to manned vehicles in which human drivers follow app directions, and unmanned vehicles following shortest path algorithms. We show that the increased usage of GPS is overall beneficial for the road network of Los Angeles, with a decrease in average travel times and total vehicle miles traveled. However, this global increased efficiency in urban mobility has negative impacts as well, such that increase in traffic in cities bordering highway, due to the rerouting of application users aiming at avoiding congestion. Those observations should lead to research on new transportation policies, mitigating the effects of app-based routing on urban areas.