3 Questions about OpenOSR – project in the «DSI Infrastructures & Labs» series
DSI Infrastructures & Labs are shareable infrastructures or structural vessels for creating collaborative research environments related to digital transformation. Prof. Dr. Manuel Günther briefly introduces OpenOSR, one of the projects in this series.
What is the main goal of OpenOSR?
With OpenOSR, on one hand we want to clarify that machine learning methods cannot be applied to every type of scenario. On the other hand, we aim to enable researchers at the University of Zurich and beyond to use our methods to improve the classification in the presence of the unknown by applying them to their own problem settings.
What is meant by «unknown» and how does OpenORS recognize it?
Machine learning algorithms are often designed to work in constraint environments. All inputs that go beyond such environments are unknown to the system. There are generally two ways to detect such unexpected inputs: For existing models, we can compute distributions of the internal representations of known classes, and declare deviations from these as unknown. New models can be trained so that the internal representations are better suited for computing such distributions.
What benefits do you expect for researchers or developers in the long term?
We hope that the research community will adopt and further develop OpenOSR in order to provide more and better methods for the detection of unexpected inputs. Due to the strict focus on reproducibility, it is possible to quickly determine whether methods are effective, and which of the methods is best suited for the specific problem settings of the researchers.
Learn more about OpenOSR here.
All projects of the series «DSI Infrastructures & Labs» can be found here.
Since July 2020, Prof. Dr. Manuel Günther is DSI Professor and Assistant Professor for Artificial Intelligence (AI) and Machine Learning at the Department of Informatics at the University of Zurich (UZH). There, he studies topics around the area of Deep Learning, which he often applies to images. Particular projects involve, for example, the identification of people based on their facial images, the development of methods to reduce bias in Deep Learning models, the extension of image-based explainable AI, the integration of insights of the visual system in the human brain into Deep Learning models, as well as the recognition of the unknown.