Uncertainty-Wizard: Fast and User-Friendly Neural Network Uncertainty Quantification
Date: 7 April 2022 / 16:30 - 17:30
Speaker: Michael Weiss
Abstract: Uncertainty and confidence have been shown to be useful metrics in a wide variety of techniques proposed for deep learning testing, including test data selection and system supervision. We present uncertainty-wizard, a tool that allows to quantify such uncertainty and confidence in artificial neural networks. It is built on top of the industry-leading tensorflow.keras deep learning API and it provides a near-transparent and easy to understand interface. At the same time, it includes major performance optimizations that we benchmarked on two different machines and different configurations. The paper describing uncertainty-wizard has won a best paper award at ICST 2021 and the tool has since raised interest from both industrical and academic users.
Biography: Michael is a Phd student at the Software Institute, Università della Svizzera italiana (USI). He co-authored eight papers in artificial intelligence and software engineering domains at various venues and journals, amongst which are IJCAI, ICSE, AAMAS and EMSE. His current research focuses on designing novel fail-safe approaches for machine learning based systems, as well as on the development of tools to facilitate the use of such approaches for developers. Before his PhD, Michael worked as a Senior Software Engineer and Technical Project Lead at a Swiss software company in the banking sector. In his free time, he likes to work on open- and closed-source side-projects. Examples include docstr-coverage (downloaded more than 200,000 times) and licenseplate (downloaded by only himself).
Chair: Rosalia Tufano
In February 2019, the Software Institute started its SI Seminar Series. Every Thursday afternoon, a researcher of the Institute will publicly give a short talk on a software engineering argument of their choice. Examples include, but are not limited to novel interesting papers, seminal papers, personal research overview, discussion of preliminary research ideas, tutorials, and small experiments.