Virtual Talk Series 2022

Talk 1

Infinitesimal Perturbation Analysis for the perimeter control problem in a Two-Region Urban Transportation Network utilizing a Macroscopic Fundamental Diagram

Abstract:

Traffic congestion in big cities has been proven to be a di cult problem with sub-optimal effects in terms of driver delay, frustration cost, and impact on the environment. In principle, many transportation networks lack a unified framework, which will coordinate the traffic in such a manner, aiming to suppress the congestion, and improving the travel time of the users commuting inside it. In view of this inadequacy, we consider the perimeter control problem for a two urban region abstracted as a stochastic hybrid system (SHS). The dynamics of each region are dictated by a Macroscopic Fundamental Diagram (MFD), in which the behavior of the system is partitioned into four different modes of operation, based on the value of density affiliated with each region situated above or below its critical density. We employ a framework known as Infinitesimal Perturbation Analysis (IPA) to dynamically adjust the controllable parameters of interest, by deriving online gradient estimators of a cost metric (i.e. maximization of the throughput of vehicles). Essentially, the SHS setting transcribes the network into a Discrete Event System (DES). The power of IPA lies within its data-driven nature to identify discontinuities in the state dynamics which subsequently signify the occurrence of an event. These estimators are used to adjust the control vector of interest iteratively through a stochastic gradient-based algorithm to maximize the system's performance. Treating the critical density of each region as a known parameter, through this framework, we aim to pinpoint the percentage of flow that traverses the "ego" region with destination to its adjacent one and vice versa to optimize the cumulative portion of vehicles that exits from each region. In comparison with other approaches, where brute force algorithms are employed to designate the optimal values of the control parameters, (IPA) anticipates the impact of events utilizing the "perturbation propagation" term and provides an estimate of the gradient without having prior knowledge of the underlying graph of the system. The novel feature in this work is the modeling architecture perceived, enabling us to capture the "perturbation propagation" when non-linearity between the interconnected regions arises due to the capacity constraints activated at the boundaries of the two regions.

Talk 2

Obfuscation and Inference: Enforcing Opacity and Utility

Abstract:

With growing concerns over the privacy of networked systems, many mechanisms for enforcing opacity as a form of privacy have been developed. In particular, obfuscating a system's outputs can preserve privacy when it is unfeasible to restrict a system's behavior. However, privacy must also be balanced with the oftentimes conflicting goal of utility. Specifically, observations of the system must reveal secrets to intended recipients while hiding them from unintended ones. In this talk, I will discuss a new framework for the obfuscation of discrete event systems and inference of intended recipients. Modeling these processes as a distributed system, I will show how utility and notions of opacity, including K-step opacity, can be formulated as temporal specifications.

Leveraging techniques from distributed reactive synthesis, I will then present how both an obfuscator for the system and inference process used by the intended recipients can be synthesized..

Talk 2

Applications of Supervisory Control to Secure System Design and Attack Synthesis

Abstract

As systems have been evolving larger and more networked, it becomes increasingly crucial to take system security into account. Due to the proliferation of different system architectures and heterogenous networks, however, keeping systems secure and designing robust protocols against attacks can be significantly challenging. From the side of system administrators, one of their central responsibilities is for preventing secret information in the system from being unintentionally exposed. For example, sensitive information stored in the system such as phone numbers, credit card transactions, passwords, or privileges (e.g., root user in Linux), should not be available freely to external agents. Besides, it is important to minimize the associated cost of implementing protections, including a decrease of usability, while the system achieves a required security level. In this talk, I first introduce a DES-based approach to model this security problem and a methodology to find a protection policy that help administrators figure out which transitions in the system must be protected. From the side of attackers, on the other hand, insecure channels/networks can be typical targets for attackers. Such an attack scheme is called Person-In-The-Middle attack. Moreover, since networked systems tend to rely on standardized protocols, vulnerabilities or misuses of protocols can be a cause of serious security incidents. Considering that the attackers’ goal is to lead the system to a violation of required properties of protocols, I will present how to synthesize attack strategies so that the attacker always eventually wins, to examine whether attackers can exploit vulnerable communication protocols. This problem of attack synthesis can be solved by leveraging existing techniques of supervisory control, and that methodology will be demonstrated with one of the most common communication protocols today: Transmission Control Protocol (TCP), specifically the phase of connection establishment using three-way handshake.