Giamattei, Luca (2024) Reasoning-Based Software Testing. [Tesi di dottorato]

Documento PDF ITEE_PHD_Thesis_Giamattei.pdf Visibile a [TBR] Amministratori dell'archivio Download (4MB) | Richiedi una copia

Abstract

With software systems becoming increasingly pervasive and autonomous, our ability to test for their quality is severely challenged. Many modern systems operate in uncertain, highly-changing environments, often required to make informed and intelligent decisions autonomously. This results in an intractable state space to explore at testing time. The state-of-the-art techniques try to keep pace, e.g., by augmenting the tester’s intuition with some form of (explicit or implicit) learning from observations to search this space efficiently. For instance, they exploit historical data to drive the search (e.g., ML-driven testing) or test execution data itself (e.g., adaptive or search-based testing). Even these data-driven techniques fall short when predicting system behavior in unobserved conditions. Despite clear advances, the need for smarter search in such a vast space keeps pressing. To overcome current software testing limitations, Reasoning-Based Software Testing (RBST) is proposed, a testing methodology reformulating software testing as causal reasoning tasks. RBST innovates software testing by shifting to a reason-driven paradigm, thanks to Causal Reasoning, contributing to the “Causal Revolution” of Judea Pearl (2011 Turing Award recipient). It constitutes a conceptual leap in how machines and humans cooperate to explore the huge search space and derive tests intelligently. Machines should support and enhance human reasoning far beyond merely identifying patterns in past data. RBST aims to emulate human-like decision-making to "intelligently" navigate the testing space. Leveraging advanced causal discovery and inference techniques, RBST moves beyond traditional ML approaches, enabling more predictive, hypothesis-driven testing. RBST is applied in both stateless and stateful testing scenarios using autonomous driving systems as a case study. Results suggest it has the potential to transform testing practices.

Downloads

Actions (login required)

Modifica documento