Projekte pro Jahr
Abstract
In this paper, we propose a novel application of syntax-guided synthesis to find symbolic representations of a model’s decision-making process, designed for easy comprehension and validation by humans. Our approach takes input-output samples from complex machine learning models, such as deep neural networks, and automatically derives interpretable mimic programs. A mimic program precisely imitates the behavior of an opaque model over the provided data. We discuss various types of grammars that are well-suited for computing mimic programs for tabular and image input data. Our experiments demonstrate the potential of the proposed method: wesuccessfully synthesized mimic programs for neural networks trained on the MNIST and the Pima Indians diabetes data sets. All experiments were performed using the SMT-based cvc5 synthesis tool.
Originalsprache | englisch |
---|---|
Titel | Bridging the Gap Between AI and Reality |
Untertitel | First International Conference, AISoLA 2023, Crete, Greece, October 23–28, 2023, Proceedings |
Herausgeber (Verlag) | Springer |
Seiten | 119-137 |
ISBN (Print) | 978-3-031-46001-2 |
DOIs | |
Publikationsstatus | Veröffentlicht - 2023 |
Veranstaltung | 1st International Conference on Bridging the Gap between AI and Reality: AISoLA 2023 - Crete, Griechenland Dauer: 23 Okt. 2023 → 28 Okt. 2023 |
Publikationsreihe
Name | Lecture Notes in Computer Science |
---|---|
Band | 14380 |
Konferenz
Konferenz | 1st International Conference on Bridging the Gap between AI and Reality |
---|---|
Kurztitel | AISoLA 2023 |
Land/Gebiet | Griechenland |
Ort | Crete |
Zeitraum | 23/10/23 → 28/10/23 |
Fingerprint
Untersuchen Sie die Forschungsthemen von „Formal XAI via Syntax-Guided Synthesis“. Zusammen bilden sie einen einzigartigen Fingerprint.Projekte
- 1 Abgeschlossen
-
EU - FOCETA - Grundlagen für kontinierliches Engineering von vertraunswertiger Autonomie
1/10/20 → 30/09/23
Projekt: Forschungsprojekt