Presentation of the meeting (PDF)

Recording of the event

Opening

Ronald Maier, Vice-Rector Digitalisation and Knowledge Transfer

Ulf Busch, CIO

Artificial Intelligence, powered by Azure

Introducing Azure Services at the University of Vienna

A brief introduction to Microsoft Azure at the University of Vienna. Furthermore, the presentation of the funding programme Funding for research with Azure services will serve as introduction for the following presentations.

Speaker: Fabian Jusufi

Decoding Odor Perception with Attention-Based Graph Neural Networks through Self-Supervised Learning

Odor perception represents a complex challenge in biological sciences, intricately linked to the interaction of molecules with the human olfactory system, comprising over 400 olfactory receptors (ORs). The multifaceted nature of these interactions, where a single molecule may bind to multiple receptors, creates a complex information flow that is ultimately interpreted as specific odors by the brain. The lack of crystal structures for all 400 ORs further complicates the understanding and prediction of these molecular interactions. In this study, we harness the potential of E3 equivariant graph neural networks (GNNs) in a novel approach to decode the molecular basis of odor perception. Initially, we pre-train our model in a self-supervised manner on a vast dataset of more than 1.2 million molecules, leveraging the inherent structural and physicochemical properties that govern their interaction with ORs. Subsequently, we fine-tune this model on a curated dataset of over 5,000 molecules, each associated with specific odor perceptions. This process yields an innovative "odor-map," a high-dimensional representation that clusters molecules based on their perceived odors. Our findings not only demonstrate the feasibility of predicting odor perceptions from molecular structures but also pave the way for identifying the molecular features responsible for specific smells, despite the absence of comprehensive structural data on ORs. This research offers profound insights into the molecular underpinnings of olfaction and presents a promising avenue for the exploration of olfactory receptor interactions and the prediction of odor characteristics.

Speakers: Oliver WiederChristian J. Binder 

Using Azure Cloud Services in the FactCheck Project

FactCheck is an ongoing research project of the Research Group Multimedia Information Systems, Faculty of Computer Science, that aims to signal conflicts within data available on the Web. This information on conflicts, which may be available in text (e.g., paragraphs in an HTML document) or multimedia form (e.g., news segments in a video), shall be extracted using a combination of approaches from the Semantic Web (e.g., structured data) and state-of-the-art AI technologies and concepts (e.g., named entity recognition or entity linking). The comparison processes for this information will be partially driven by human intelligence and human feedback, which is why approaches for user identities and user management (e.g., Azure Entra ID) will also be investigated. For the deployment of the FactCheck prototype(s), a hybrid approach is considered, which allows for the use of both scalable Azure services (e.g., cognitive services like AI Video Indexer or user management) as well as available on-premises infrastructure (e.g., VMs or databases) at the University of Vienna to achieve suitable tradeoffs in terms of security, privacy, and costs. To keep the deployment highly flexible and modular, parts of this deployment may be containerized, thus simplifying deployment on both Azure and local infrastructure.

Speakers: Wolfgang Klas, Mara Sophie Aichinger 

ZID Services and AI 

Data Enrichment by AI: UNIDAM as an Example

With the development of powerful AI and the rise in computing power researchers are handed tools to sift through ten thousands of pages or pictures and extract substantial information by the bundle. This information can then be used to enrich the data and make searches more meaningful and exact, but also initiate unthought-of research angles. UNIDAM is a Digital Asset Management system run by PHAIDRA services and used for picture objects mainly by the humanities. The largest section is comprised by the History of Arts department which uses UNIDAM as research database, but also as a tool to quickly create Powerpoint presentations of pictures to be shown in teaching. At the moment UNIDAM holds half a million objects, so it is well worth to try enrichment by automated or half-automated processes. Though this goal is not yet reached by far, Martin, János and their teams have begun to include several enrichment sources: Named Entity Recognition (to link artist names to Wikidata, GND or VIAF), Topic Modelling (to comprise new perspectives on voluminious text data) or correction of OCR and HCR results. As an example for such an automated enrichment they present their current efforts to enrich a small section of content in UNIDAM by a plugin using the university's Azure scripting backend. That way they can harness the power of the latest OpenAI large language model with picture description ability, and retrieve more or less meaningful so-called Iconclass codes. Those codes then can be used to search and classify the pictures in UNIDAM by their iconographic content. They will describe difficulties, procedures and learnings they identified in the course of this project as well as future prospects and general considerations.

Speakers: Martin Gasteiner (unable to attend), János Békési

News and Feedback

Michaela Bociurko, IT Communications and Marketing / ZID

Outlook

Ronald Maier, Vice-Rector Digitalisation and Knowledge Transfer


Moderation:
Michaela Bociurko, IT Communications and Marketing / ZID
Sara Curtis, IT Communications and Marketing / ZID

  • No labels