Reasoning under partial observability in heterogeneous networked systems

Networked systems are collections of interconnected devices and software components that communicate to provide coordinated functionality, such as sensing, control, data processing, and access enforcement. Many everyday services rely on such systems, including building automation, cloud applications, industrial monitoring, and connected medical or sensor devices. In these systems, overall behavior is shaped by configuration choices, such as access rules, control logic, and communication settings, and by how these choices interact across components. Small configuration mistakes or unintended rule interactions can therefore lead to serious safety or security problems. These systems are often assembled from components developed independently and are not always fully observable during operation, which makes it difficult to determine whether the system as a whole behaves correctly and safely. This thesis addresses the problem of reasoning about correctness, safety, and redundancy in configured networked systems under uncertainty. It argues that this problem can be addressed by introducing a common, easy-to-use semantic layer into which heterogeneous configurations can be compiled, and which explicitly supports the representation of incomplete information. Such a semantic layer enables sound, system-wide reasoning over the set of possible behaviors consistent with what is known about the system. To realize this approach, the thesis leverages conditional tables, a database-inspired representation for incomplete information. Conditional tables provide a unifying symbolic abstraction for modeling heterogeneous data and control logic while explicitly encoding uncertainty. Within this framework, properties such as safety, policy compliance, and reachability can be expressed as declarative queries and evaluated soundly under partial observability. The same representation supports semantic minimization, allowing complex systems to be reduced to smaller models that preserve behaviors of interest. A prototype implementation demonstrates that conditional-table–based reasoning is practical and scales to realistic networked scenarios under typical deployment sizes.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois Urbana-Champaign

Year dc:date

2025

Author dc:creator

• Anwar, Mubashir

Contributors dc:contributor

• Caesar, Matthew

• Heterogeneous networked systems
• Reasoning under uncertainty
• Declarative verification

Statement dc:rights

• Copyright 2025 Mubashir Anwar

Language dc:language

en

Handle dc:identifier

https://hdl.handle.net/2142/132709

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/132709