Software engineering has a success history of evolving symbolic techniques, e.g., formal methods and programming languages, to solve increasingly challenging problems like providing safety and performance guarantees for autonomous intelligent systems fulfilling mission-critical functions. With the availability of machine learning (ML) techniques, software engineering expanded its set of problems to how learning from data enables applications from code summarization & generation to automatic program repair & formal verification. The integration of symbolic and ML techniques has opened new novel methodological challenges that go beyond applying ML to build software (ML4SE) or applying software engineering to build ML (SE4ML). These challenges fall under the umbrella of Neuro-Symbolic methods and comprise problems of “how to reason about learning” and “how to learn about reasoning”.
The NSE workshop aims to discuss these problems in the context of software engineering tasks that have been transformed by the adoption of machine learning techniques. We invite insights on merging symbolic and ML techniques across the software development life-cycle, its activities, tasks, and tools. We welcome case studies, conceptual innovative approach descriptions, empirical research, and more formal or theoretical considerations.
Our goal is to collect experiences, challenges, and solutions involved in combining symbolic methods and machine learning to tackle new and traditional challenges of software engineering tasks from requirements to analysis & design, coding, testing, and maintenance & evolution. We welcome contributions in any of the following formats:
Full research papers
Case studies
Proofs-of-concept
New ideas and emerging results
Evaluation of tools
Controlled experiment reports
Plenary
Hide plenary sessions
This program is tentative and subject to change.
Program Display Configuration
You're viewing the program in a time zone which is different from your device's time zone change time zone
Sat 3 May
Displayed time zone: Eastern Time (US & Canada)change
The integration of symbolic and ML techniques has opened new novel methodological challenges that go beyond applying Machine Learning (ML) to build software (ML4SE) or applying software engineering to build ML (SE4ML). These challenges fall under the umbrella of Neuro-Symbolic methods and comprise problems of “how to reason about learning” and “how to learn about reasoning”. NSE aims to discuss these problems in the context of software engineering tasks, which in turn have been the subject of innovation through the adoption of machine learning techniques. To illustrate that, we selected a broad non-exclusive list of topics (see below). Ultimately, this workshop invites insights on merging symbolic and ML techniques across the software development life-cycle, its activities, tasks, and tools. We welcome case studies, conceptual innovative approach descriptions, and empirical research which explore how formal reasoning can benefit from learning from data, e.g., trimming combinatorial search spaces, reconfiguring symbolic representations, generating new rules, etc. Conversely, we look forward to contributions that discuss how learning from data can be improved by being steered by reasoning, e.g., regularization, shaping goals/rewards, ensemble learning, etc. If you are also excited about any of these perspectives, please join us in shaping the future of software engineering.
NSE seeks submissions describing novel research, emerging ideas, and work-in-progress describing original and unpublished results in the field of Neuro-symbolic methods for software engineering.
Neuro-symbolic methods in automated software engineering tools, e.g., code & test generation, bug fixing, code summarization, code review, etc.
Neuro-Symbolic agents to support collaboration and decision making in software teams.
Neuro-Symbolic methods in validation and verification tools.
Neuro-Symbolic methods for designing safety-mission-critical systems.
Neuro-Symbolic methods for extracting and maintaining knowledge graphs for software engineering.
Methods for reasoning about learning from software data.
Methods for learning while reasoning about software, e.g., automatically & adaptively determining decision thresholds and magnitude of actions for a desired effect of a software tool & technique.
Methods for applying prior symbolic or probabilistic knowledge to new or improved software tools & methods.
Submissions must conform to the IEEE conference proceedings template, specified in the IEEE Conference Proceedings Formatting Guidelines, using the double-column format. Research papers must have a maximum length of 6 pages including references. Short papers are limited to 4 pages including references. Industry demonstrations and extended abstracts must have a maximum length of 2 pages including references. There is no limit on the number of submissions an author may submit. We will follow a single-blind process, i.e., anonymizing the submission is not required.
We require all submissions to be original, i.e., they should not have been previously published in any conference proceedings, book, or journal and should not currently be under review for another archival conference. Papers must be submitted electronically by the workshop deadline via the submission link (see above).
Keynotes
Asim Munawar
Reasoning Revolution: Cracking the Code of LLM Intelligence
Abstract
Large Language Models (LLMs) have brought significant advancements in artificial intelligence, showing capabilities that were once thought impossible. This talk examines how these models reason, process information, and solve problems. We will discuss the developments that allow LLMs to simulate reasoning, their limitations, and the challenges of aligning their outputs with human expectations. Using practical examples and recent research, attendees will gain a better understanding of how LLMs work and how to use their reasoning abilities effectively. This session is designed for anyone interested in understanding and utilizing these powerful tools.
Biography
Dr. Asim Munawar is a Project Lead at the IBM Watson Research Center in New York, where he is at the forefront of advancing IBM's in-house LLMs for enhanced reasoning and planning capabilities. With a distinguished 12-year tenure at IBM Research, Dr. Munawar has held pivotal roles, including Manager and Program Director for Neuro-Symbolic AI. He earned his Ph.D. from Hokkaido University and has since led multiple groundbreaking research initiatives in deep learning. His primary focus is on pioneering next-generation AI by integrating symbolic reasoning with machine learning techniques. Dr. Munawar has authored over 60 publications in leading journals and peer-reviewed conferences. He also serves as a Board member for Packt Publishing India and the National Center of Artificial Intelligence (NCAI) in Pakistan.
Stefan Szeider
Neural Meets Symbolic: Synergies Between Language Models and Constraint Reasoning
Abstract
Integrating Large Language Models (LLMs) with traditional solving techniques creates new synergies in automated reasoning. This talk explores both (i) how LLMs can enhance SAT and constraint solving through structural analysis and search guidance and (ii) how formal reasoning can help LLMs tackle hard reasoning and optimization problems. We will present case studies exploring the practical advances and future potential of combining neural and symbolic approaches in computational reasoning.
Biography
Stefan Szeider is a Professor and Chair of the Algorithms and Complexity Group at TU Wien, Austria, and a visiting scientist at the Simons Institute for the Theory of Computing at UC Berkeley. As a key researcher in the Austrian cluster of excellence Bilateral AI, he explores new neurosymbolic approaches to algorithmic challenges in AI and Explainability. Following his doctorate in Mathematics from the University of Vienna, Dr. Szeider has held research positions at universities in the UK and Canada and received an ERC Starting Grant. He co-founded the Vienna Center for Logic and Algorithms and continues to develop rigorous approaches to AI reasoning and explanation.
