Alexandru Constantin Serban's research while affiliated with Radboud University and other places

Publications (5)

Preprint
We provide a complete characterisation of the phenomenon of adversarial examples - inputs intentionally crafted to fool machine learning models. We aim to cover all the important concerns in this field of study: (1) the conjectures on the existence of adversarial examples, (2) the security, safety and robustness implications, (3) the methods used t...
Chapter
This paper analyses security aspects of the ETSI ITS standard for co-operative transport systems, where cars communicate with each other (V2V) and with the roadside (V2I) to improve traffic safety and make more efficient use of the road system. We focus on the initial information exchange between vehicles and the road side infrastructure responsibl...

Citations

... Failure recovery in micro-service architectures Jin et al. (2020) Hard to achieve heterogeneous redundancy Serban (2019) Reuse in ML-based systems is hard Kusmenko et al. (2019) Lack of knowledge required to build ML-based systems Muzaffar et al. (2015), I4 ...
... The BSM is a twopart message -the first (default) part is periodic (sent at a rate maximum rate of 10 Hz) and the second part is event-driven (e.g., for emergency braking, traffic jams, etc.) and included in the next periodic BSM message. The C-ITS equivalent of BSM are the periodic cooperative awareness message (CAM) and the (event-driven) decentralized environmental notification message (DENM) [21]. The eventdriven BSM messages are suitable for local neighborhoods (e.g., single hop broadcast) where DENMs can be used for specific geographical areas (e.g., multiple hops geocast). ...
... They further map components to these different categories (e.g., Sensing as part of Perception) and compare their architecture to the Observe-Orient-Decide-Act (OODA) model [16], which can be applied to discretize a human driver. Serban et al. [100] provide another functional software architecture for autonomous vehicles. They cluster different multiple functional components to the classes Sensors Abstraction, Data Management, Actuators Interface, Sensor Fusion, World Model, Behavior Generation, Planning, Vehicle Control and System and Safety Management. ...
... Given these considerations, we conclude that the issue of safety validation cannot be 'solved', as in proven mathematically with absolute certainty, at least without assuming numerous conditions such as sun illumination, other agents, perfect road and so on. Junietz et al. [134] and Serban et al. [135] provide an evaluation of the current safety validation methods. As stated in [135], the current ISO standard for intelligent vehicles, ISO 26262, fails to cover emergent concerns related to autonomous decisions (i.e., path planning). ...