Risk-averse Selfish Routing

on YouTube ↗32m

" summary

Models routing as a nonatomic flow with edge delays plus independent noise and users minimizing a mean+variance objective with common risk coefficient R. With a bounded variance-to-mean ratio K the paper defines the price of risk aversion = 1+eta R K, where eta is a topology parameter (eta=1 for series-parallel networks, eta=2 for the Braess alternation). The bound is tight via inductive graph constructions, and preference heterogeneity helps equilibrium cost only on series-parallel networks and can hurt otherwise.

AI-written summary · about

" chapters (16)

1. 0:00Intro
2. 0:34Traffic conditions are uncertain
3. 1:41Commuters pad travel times
4. 3:10Goal
5. 4:06Understanding traffic congestion
6. 5:36Price of anarchy = 4/3
7. 8:15Risk sensitivity of price of anarchy
8. 10:04Routing games with stochastic delays
9. 12:15Risk-averse vs Risk-neutral Equilibrium
10. 14:04Equilibrium characterization for mean-stdev risk
11. 15:05Are Risk-Averse Equilibria Efficient?
12. 18:45Risk-averse vs Risk-neutral equilibria
13. 21:09Price of Risk Aversion (PRA)
14. 21:53Price of Risk Aversion: Upper Bound for Arbitrary Latency Functions
15. 26:57Heterogeneous players
16. 28:57Summary