Provably-Safe Offloading of Neural Network Controllers for Energy-Efficiency in Autonomous Driving Systems

DATE/TIME: Wednesday, January 23rd, 12:00pm
SPEAKER: Mohanad Odema
To mitigate the high energy demand of Neural Network (NN) based Autonomous Driving Systems (ADSs), we consider the problem of offloading NN controllers from the ADS to nearby edge-computing infrastructure, but in such a way that formal vehicle safety properties are guaranteed. In particular, we propose the EnergyShield framework, which repurposes a controller "shield" as a low-power runtime safety monitor for the ADS vehicle.

Fairness-Aware learning over Graphs

DATE/TIME: Tuesday, April 2nd, 12:00pm
SPEAKER: Oyku Deniz Kose
While graph-based ML models nicely integrate the nodal data with the connectivity, they also inherit potential unfairness. Using such ML models may therefore result in inevitable unfair results in various decision- and policy-making in the related applications. While fairness and explainability have attracted increasing attention in responsible ML, they are mostly under-explored in the graph domain.