Nicolas S. Schuler

Publications

Download all citations (.bib)

Selected papers

Doctoral research

When a learned model passes a prediction to a planner or rule-based component, that component needs to know which assumptions it can rely on. This doctoral research proposal sets out contracts and runtime checks at those boundaries, with an evaluation plan for studying dependable adaptation as components and requirements change.

Architecting Self-Adaptive Systems with Learned and Symbolic Components · Research proposal, accepted at the ECSA 2026 Doctoral Symposium, where it received the ⭐ Best Paper Award. Full citation.

The proposal builds on a position paper that argues for composable AI systems: instead of one monolithic model, learned and symbolic parts meet at deliberate seams that expose typed boundary objects, versioned constraints, and decision traces.

Beyond Monolithic Models: Symbolic Seams for Composable Neuro-Symbolic Architectures · Position paper, arXiv preprint. Full citation.

Reasoning in software

Two studies look at what happens when language models and logic take part in software reasoning. One measures whether language models preserve how strongly a stakeholder meant a requirement when extracting it, and whether their confidence signals reveal when they do not. The other shows how an image classifier’s heatmap explanation can be translated into logical facts that a probabilistic reasoner can check.

When Weak Intent Becomes a Requirement: Limits of Uncertainty Signals in LLM-Assisted Requirements Engineering · Empirical study, submitted to the Journal of Systems and Software. Full citation.

Bridging Explanations and Logics: Opportunities for Multimodal Language Models · Position paper, accepted at AISoLA 2025. Full citation.

Further studies

TraceGate controls which crash details and diagnostic helpers are available to a language-model-based repair system, so that investigating a failure becomes part of the repair strategy. It recovered additional correct repairs across the evaluated benchmarks and model sizes, although with additional repair budget rather than at equal cost.

TraceGate: Policy-Based Disclosure of Structured Crash Evidence for LLM-Assisted Debugging · Empirical study, accepted at the ISSRE 2026 Research Track. Full citation. Project page.

When models learn from data held by separate participants, how useful are the explanations of their predictions? This study evaluates and combines attribution methods, finding improvements in explanation metrics; a complementary user study shows that higher scores do not consistently mean explanations people prefer.

Metric-Guided Attribution Fusion in Explainable Federated Learning: An Empirical Evaluation · Empirical study, preprint submitted to Information Fusion. Full citation. Project page.

Publication list

Dates are labelled as conference dates, preprint postings, artefact releases, or thesis dates. Acceptance and manuscript availability are listed separately: a paper accepted at a venue may still link to a preprint.

PhD research

  1. Conference dates
    –
    TraceGate: A Framework for Policy-Controlled Failure Evidence in LLM-Assisted Program Repair (conference abstract)
    Accepted · ASE 2026 Tools and Datasets
    Nicolas S. Schuler, Vincenzo Scotti, and Raffaela Mirandola
    Show BibTeX 
    @inproceedings{schuler2026tracegatetool,
      author    = {Schuler, Nicolas S. and Scotti, Vincenzo and Mirandola, Raffaela},
      title     = {{TraceGate}: A Framework for Policy-Controlled Failure Evidence in {LLM}-Assisted Program Repair},
      booktitle = {Proceedings of the 41st IEEE/ACM International Conference on Automated Software Engineering (ASE 2026)},
      year      = {2026},
      doi       = {10.1145/3832783.3834596},
      note      = {Tools and Datasets track. To appear}
    }
  2. Replication package released
    When Weak Intent Becomes a Requirement: Limits of Uncertainty Signals in LLM-Assisted Requirements Engineering
    Submitted · Journal of Systems and Software
    Nicolas S. Schuler, Vincenzo Scotti, and Raffaela Mirandola
    Show BibTeX 
    @unpublished{schuler2026weakintent,
      author = {Schuler, Nicolas S. and Scotti, Vincenzo and Mirandola, Raffaela},
      title  = {When Weak Intent Becomes a Requirement: Limits of Uncertainty Signals in {LLM}-Assisted Requirements Engineering},
      year   = {2026},
      note   = {Submitted to the Journal of Systems and Software. Replication package: \url{https://doi.org/10.5281/zenodo.22802187}}
    }
  3. Preprint posted
    TraceGate: Policy-Based Disclosure of Structured Crash Evidence for LLM-Assisted Debugging DOI link
    Accepted · ISSRE 2026 Research Track · KIT preprint
    Nicolas S. Schuler, Balint Mate, Vincenzo Scotti, and Raffaela Mirandola
    Show BibTeX 
    @inproceedings{schuler2026tracegate,
      author    = {Schuler, Nicolas S. and Mate, Balint and Scotti, Vincenzo and Mirandola, Raffaela},
      title     = {{TraceGate}: Policy-Based Disclosure of Structured Crash Evidence for {LLM}-Assisted Debugging},
      booktitle = {Proceedings of the 37th IEEE International Symposium on Software Reliability Engineering (ISSRE 2026)},
      publisher = {IEEE},
      year      = {2026},
      note      = {Research track. To appear. Preprint: \url{https://doi.org/10.5445/IR/1000195694}}
    }
  4. Preprint posted
    Metric-Guided Attribution Fusion in Explainable Federated Learning: An Empirical Evaluation DOI link
    Preprint · Submitted to Information Fusion · SSRN
    Nicolas S. Schuler, Matteo Camilli, Vincenzo Scotti, and Raffaela Mirandola
    Show BibTeX 
    @misc{schuler2026attribution,
      author       = {Schuler, Nicolas S. and Camilli, Matteo and Scotti, Vincenzo and Mirandola, Raffaela},
      title        = {Metric-Guided Attribution Fusion in Explainable Federated Learning: An Empirical Evaluation},
      howpublished = {SSRN preprint},
      year         = {2026},
      doi          = {10.2139/ssrn.7129255},
      note         = {Submitted to Information Fusion}
    }
  5. Preprint posted
    Architecting Self-Adaptive Systems with Learned and Symbolic Components DOI link
    Accepted · ECSA 2026 Doctoral Symposium · KIT preprint
    Best Paper Award
    Nicolas S. Schuler
    Show BibTeX 
    @inproceedings{schuler2026architecting,
      author    = {Schuler, Nicolas S.},
      title     = {Architecting Self-Adaptive Systems with Learned and Symbolic Components},
      booktitle = {Software Architecture: ECSA 2026 Tracks and Workshops},
      publisher = {Springer},
      year      = {2026},
      note      = {Doctoral Symposium. To appear. Preprint: \url{https://doi.org/10.5445/IR/1000195259}}
    }
  6. Preprint submitted
    Beyond Monolithic Models: Symbolic Seams for Composable Neuro-Symbolic Architectures DOI link
    Preprint · arXiv
    Nicolas S. Schuler, Vincenzo Scotti, and Raffaela Mirandola
    Show BibTeX 
    @misc{schuler2026seams,
      author        = {Schuler, Nicolas S. and Scotti, Vincenzo and Mirandola, Raffaela},
      title         = {Beyond Monolithic Models: Symbolic Seams for Composable Neuro-Symbolic Architectures},
      year          = {2026},
      eprint        = {2603.15087},
      archivePrefix = {arXiv},
      primaryClass  = {cs.SE},
      doi           = {10.48550/arXiv.2603.15087}
    }
  7. Preprint posted
    Bridging Explanations and Logics: Opportunities for Multimodal Language Models DOI link
    Accepted · AISoLA 2025, Formal Methods for Intersymbolic AI · KIT preprint
    Nicolas S. Schuler, Vincenzo Scotti, Matteo Camilli, and Raffaela Mirandola
    Show BibTeX 
    @inproceedings{schuler2025bridging,
      author    = {Schuler, Nicolas S. and Scotti, Vincenzo and Camilli, Matteo and Mirandola, Raffaela},
      title     = {Bridging Explanations and Logics: Opportunities for Multimodal Language Models},
      booktitle = {Bridging the Gap Between AI and Reality (AISoLA 2025)},
      publisher = {Springer},
      year      = {2025},
      note      = {Formal Methods for Intersymbolic AI track. Preprint: \url{https://doi.org/10.5445/IR/1000184717}}
    }

Theses before the PhD

  1. Thesis period ended
    Master's Thesis: Integrating Explainability into Federated Learning: A Non-functional Requirement Perspective (PDF, 17 MB) (minisign signature)
    – at the Self-adaptive Software-intensive Systems (SASIS) group, KIT
    Grade: 1.0 (German scale; 1.0 is the best grade)
    Show BibTeX 
    @mastersthesis{schuler2025thesis,
          author = {Schuler, Nicolas S.},
          title  = {Integrating Explainability into Federated Learning: A Non-functional Requirement Perspective},
          school = {Karlsruhe Institute of Technology (KIT)},
          year   = {2025},
          url    = {https://n-schuler.dev/media/docs/ma-thesis.pdf}
        }
  2. Repository posting
    Bachelor's Thesis: Verteilte Nutzungskontrolle und Provenance Tracking am Beispiel von Cloud-Technologien DOI link
    – in cooperation with the Fraunhofer IOSB, KIT
    Grade: 1.3
    Show BibTeX 
    @mastersthesis{schuler2020thesis,
          author = {Schuler, Nicolas S.},
          title  = {Verteilte Nutzungskontrolle und Provenance Tracking am Beispiel von Cloud-Technologien},
          school = {Karlsruhe Institute of Technology (KIT)},
          type   = {Bachelor's thesis},
          year   = {2020},
          doi    = {10.5445/IR/1000128308}
        }