Workshops and Tutorials at PACT 2026

1st Workshop on ML for Assisting Code Quality (MLAC)

Scheduling: Monday, October 19 (morning)

Workshop Chair

Jay Lofstead, Sandia National Laboratories

Program Committee

Jim Willenbring (SNL)
Vanessa Sochat (LLNL)
Ewa Deelman (ICI/USC)
Sandra Gesing (USRSE)
Weronika Filinger (EPCC)

Scope

Using ML tools, such as LLMs and coding assistants are changing the way software is developed. Vibe coding even removes the need for a user to really understand code, but instead requires guiding the model to correcting errors found in the previous generated output and adding new features incrementally. These techniques promise to accelerate software development and/or ease repetitive tasks letting software engineers focus on more complex and important code areas. Easy applications, such as simple, web based systems, can be fully generated today. More complex applications, such as science simulations using complex physics, specific data mesh designs, intricate data distribution and management, and underlying hardware specific requirements are essentially impossible. This capability range leaves many application classes potentially possible, but with unknown pitfalls for different kinds of software problems. Working to understand how to generate high quality code with ML tools is essential for productivity today and for the future of the software engineering profession.

This workshop seeks to explore how to use ML-related tools to try to support software development ultimately improving code quality. Human programming introduces errors regularly. The promise of ML-assisted programming is simple errors could be eliminated by using adapted generated code fixed by the ML tools. More subtle errors, such as security vulnerabilities, can use ML tools to do detailed analysis using all available knowledge rather than what any single researcher knows. With the potential success for generated tests, test coverage, code security, and ability to generate code to address complex problems wildly variable, gathering to share recent experiments and developments will help the entire community understand how to better integrate these tools into the software engineering process.

Main topics

ML-based code generation successes and challenges for various application domains
ML-based software testing, test coverage, and all other testing related topics
Security auditing using ML tools
Software design assistance using ML tools
Balancing the human and ML costs over the short term through the long term for software engineering and full product life cycle costs
Related topics of using ML tools to support software related activities

Submission

To better support Software Engineering professionals, we will accept abstracts for talks, in addition to solely peer-reviewed submissions.

Link to submission portal: https://easychair.org/conferences/?conf=mlac26

Bio

Jay Lofstead is a Principal Member of Technical Staff at Sandia National Laboratories. His research interests focus on large-scale data management and trustworthy scientific computing. In particular, he works on storage, IO, metadata, workflows, reproducibility, software engineering, machine learning, and operating system-level support for any of these topics. Broadly across these topics, he is also deeply interested in ethics related to these topics and computing in general and how to drive inclusivity across the computationrelated science domains. Dr. Lofstead received his Ph.D. in Computer Science from the Georgia Institute of Technology in 2010. Most recently, he co-supervised a student at St. John’s undergraduate thesis on using LLMs to generate unit tests for C++ code. He has also worked extensively with the Research Software Engineering group at Sandia, including their cautious adoption of ML tools for supporting software engineering. He has also been working with internal teams at Sandia, developing ML use policy related to sensitive and high-consequence code development using ML tools.

Back to Workshops