Auf dieser Seite werden die nächsten fünf Seminartermine angezeigt. Für weitere Termine bitte unten auf "weitere Termine" klicken oder bei Alle Termine nachschauen. Neue Terminseiten können ebenfalls auf Alle Termine angelegt werden. Kalenderdatei (alle zukünftigen Termine): iCal (Download)
Ort: Raum 010 (Gebäude 50.34)
LLM-assisted Generation of Semantic Tests for Model Transformation Language Code
| Vortragende(r)
|
Kirill Dyachenko
|
| Vortragstyp
|
Bachelorarbeit
|
| Betreuer(in)
|
Bowen Jiang
|
| Vortragssprache
|
Englisch
|
| Vortragsmodus
|
in Präsenz
|
| Kurzfassung
|
Large Language Models (LLMs) demonstrate promising capabilities in generating code for model transformation languages (MTL). However, syntactically correct transformations can exhibit erroneous behavior, necessitating semantic evaluation. In this work, the LLM4MTL pipeline is augmented with a mechanism for generating and verifying semantic tests using LLMs. The generated tests are checked for executability and compared against reference transformations before being used to evaluate the generated MTL code. Parser results and data on semantic test failures serve as feedback for the iterative refinement of the generated transformations and tests. The evaluation examines the reliability of LLM-generated tests, their ability to detect semantic errors missed by syntactic checks, and the effectiveness of feedback-based MTL code refinement across various model transformation languages.
|
Ort: Raum 348 (Gebäude 50.34)
Deriving Abstract Penetration Test Cases from LLM-Annotated Architecture Models
| Vortragende(r)
|
Thomas Riekert
|
| Vortragstyp
|
Masterarbeit
|
| Betreuer(in)
|
Mahdi Jafari
|
| Vortragssprache
|
Englisch
|
| Vortragsmodus
|
in Präsenz
|
| Kurzfassung
|
Authorization weaknesses are usually found only during or after implementation, even though many of them come from design decisions. This thesis moves part of identifying weaknesses into the design phase and derives Abstract Penetration Test Cases (APTCs) from Palladio Component Model (PCM) architectures. A security annotation adds three things to the model: which actors each operation is intended for, how far each actor is trusted, and how sensitive each component's data is. Deterministic rules then trace every path from an entry point to the data it reaches. Each path that breaks a rule against missing authentication (CWE-306) or missing authorization (CWE-862) becomes an APTC. Writing the annotation takes the most effort, so an LLM drafts it from the model and a fixed vocabulary, and an architect reviews the draft. The evaluation covers four case studies (MoPla, crAPI, CoronaWarnApp and EVerest) and an expert study. It looks at whether the annotation can be applied in practice, if the APTCs stay consistent when the model changes and if experts find them useful.
|
Ort: Raum 348 (Gebäude 50.34)
Ort: Raum 010 (Gebäude 50.34)
Ort: Raum 010 (Gebäude 50.34)
weitere Termine
