Modern Software Engineering in the AI Era

Opening & Keynote Speech: Modern SE in the AI age

Presenter: Prof. Tim Menzies

Abstract: Everyone says AI needs more: more parameters, more data, more GPUs, more money. Really? Complexity has costs, and they're not always obvious until something goes boom. In this talk, I'll walk through thirty years of building AI systems that worked because they were simple, not despite it. Like the expert system for raising pigs that outperformed the human expert who wrote its rules—built by a junior master's student (i.e. me). Or modern "compact AI" methods that match GPT-class results using 1/1000th of the data. Here's the heresy: in most real software systems, complexity collapses. Behavior funnels. The winning move isn't throwing more compute at the problem—it's asking a better question. If you've ever suspected that the AI hype machine is missing something important, come find out what.

Presenter: A/Prof. Xi Zheng

Abstract: Learning-enabled autonomous systems—such as self-driving vehicles and intelligent drones—pose unprecedented challenges for safety assurance due to the opaque and unpredictable nature of deep neural networks. This talk introduces NeuroStrata, a new neurosymbolic architecture for autonomous systems, which marks a paradigm shift from black-box learning to interpretable, reasoning-based intelligence. By integrating neural perception with symbolic reasoning, NeuroStrata enables certifiable AI, bridging the gap between data-driven adaptability and formal verifiability. This vision is now being realized through a neurosymbolic perception module deployed in collaboration with an Australian drone company, demonstrating real-world feasibility for safety-critical applications.

Presenter: Dr. Xiao Cheng

Abstract: Program analysis is fundamental to software quality assurance, as it enables automated reasoning about program behavior and the verification of properties against specifications. Large language models (LLMs) have shown great promise for program analysis due to their strong capability for semantic understanding, acquired through extensive training on large-scale codebases. However, because of the hallucination tendency of LLMs, they may produce spurious or counterfactual outputs, hurting the accuracy of program behavior reasoning. This problem becomes significantly more serious when applying LLMs to whole-program analysis, where the target is not toy, function-level code but industrial-scale codebases consisting of millions of lines of code. These limitations hinder the practical adoption of LLM-powered program analysis.

In this talk, I will present three works on LLM-powered whole-program analysis and its key applications: library API specification generation (ICSE 2026), fuzz driver enhancement (NDSS 2026), and codebase-level program repair (FSE 2026). I will elaborate on our methodologies for effectively combining LLMs with fundamental program analysis techniques and for mitigating hallucinations in LLM-powered whole-program analysis through program decomposition, verification, and retrieval-augmented generation techniques.

Presenter: Dr. Yongqiang Tian

Abstract: The field of compiler engineering has long relied on deep human expertise and painstaking manual effort for tasks like finding subtle bugs or optimizing performance. This labour-intensive process, however, fundamentally limits the pace of development and analysis. This talk presents our recent work on how Large Language Models (LLMs) can fundamentally supercharge this process, allowing engineers to operate with unprecedented speed and efficiency. We will demonstrate how LLMs can intelligently (1) generate candidates for missed peephole optimizations and (2) implement language-specific transformations to automate the debugging workflow.

Presenter: Professor Hoa Khanh Dam

Abstract: This talk envisions a transformative paradigm in software engineering, where Artificial Intelligence, embodied in fully autonomous agents, becomes the primary driver of the core software development activities. We introduce a new class of software engineering agents, empowered by Large Language Models and equipped with beliefs, desires, intentions, and memory to enable human-like reasoning. These agents collaborate with humans and other agents to design, implement, test, and deploy software systems with a level of speed, reliability, and adaptability far beyond the current software development processes. Their coordination and collaboration are governed by norms expressed as deontic modalities - commitments, obligations, prohibitions and permissions - that regulate interactions and ensure regulatory compliance. These innovations establish a scalable, transparent and trustworthy framework for future Human-AI software engineering teams.

Presenter: Dr. James Hoang

Abstract: Utilising quantum computing technology to enhance artificial intelligence systems is expected to improve training and inference times, increase robustness against noise and adversarial attacks, and reduce the number of parameters without compromising accuracy. However, moving beyond proof-of-concept or simulations to develop practical applications of these systems while ensuring high software quality faces significant challenges due to the limitations of quantum hardware and the underdeveloped knowledge base in software engineering for such systems. In this work, we have conducted a systematic mapping study to identify the challenges and solutions associated with the software architecture of quantum-enhanced artificial intelligence systems. The results of the systematic mapping study reveal several architectural patterns that describe how quantum components can be integrated into inference engines, as well as middleware patterns that facilitate communication between classical and quantum components. Each pattern realises a trade-off between various software quality attributes, such as efficiency, scalability, trainability, simplicity, portability, and deployability. The outcomes of this work have been compiled into a catalogue of architectural patterns.

Presenter: Dr. Guoxin Su

Abstract: Deep reinforcement learning (DRL) has emerged as a powerful paradigm for solving complex decision-making problems. However, DRL-based systems still face significant dependability challenges particularly in real-time environments due to the simulation-to-reality gap, out-of-distribution observations, and the critical impact of latency. Latency-induced faults, in particular, can lead to unsafe or unstable behaviour, yet existing fault-tolerance approaches to DRL systems lack formal methods to rigorously analyse and optimise performance and safety simultaneously in real-time settings. In this talk, I will present a formal framework for the design and analysis of real-time switching mechanisms between DRL agents and alternative controllers. Our framework models switching behaviour using timed automata and introduces multi-objective model checking to evaluate the switch design against both soft and hard performance requirements. A GPU-accelerated implementation of our novel method demonstrates superior scalability compared to the state-of-the-art probabilistic model checking tools.

Presenter: Dr. Hong Jin Kang

Abstract: Program analysis is essential for software correctness and security, yet its practical impact is often limited by tools that do not align with developers' reasoning processes or accommodate multiple perspectives. A human-centered approach can help close this gap.

This talk will present a multi-agent framework for vulnerability detection in which multiple LLM agents adopting different roles debate and refine vulnerability hypotheses, improving detection performance and enabling the discovery of multiple CVEs. It will also cover an interrogative debugger for taint analysis that allows developers to ask "why," "why-not," and "what-if" questions to explain unexpected or missing security warnings, reducing cognitive load and improving sensemaking. Finally, we will outline future directions for advancing program analysis tools to address long-standing challenges for effective program analysis.

Open discussion on research collaboration opportunities and future directions