Multimodal Analytics combine single-modal research branches to identify more complex interrelationships. Interdisciplinary methods enable more efficient and flexible solutions for complex problems.
The DSAI team provides data science and artificial intelligence expertise for a wide variety of modalities (i.e., text, audio, time-series, and geo-spatial data) as well as multimodal applications combining these modalities. Multimodal research tasks require coordinated collaboration and interdisciplinary communication. Each Research Topic is driven by a senior scientist or senior engineer within the DSAI team.
Language Analytics
Language Analytics focus on approaches to gain insights from human communication, taking a media holistic approach to the application of audio analysis, natural language processing, and Large Language Models (LLMs). DSAI is leading expert in the areas of fake news, disinformation campaign detection, and monitoring and audio analysis in Austria. The team applies local LLMs to information extraction, summarization, and the semi-automated construction of Knowledge Graphs. Read more
Graph Analytics
Graph Analytics concerns the application of multimodal graph analysis to transactions and interactions in both virtual and geospatial realms (covering, e.g., transaction networks in eCommerce as well as interactions in spatial networks). DSAI offer leading expertise in spatiotemporal analytics, such as analysis of trajectories from a variety of tracking systems, including GPS, cellular, and video-based, as well as in cryptoasset forensics; for example, providing anomaly detection solutions for maritime safety. Read more
Industrial Analytics
Industrial Analytics concerns the application of artificial intelligence in industrial and production environments tackling predictive maintenance, anomaly detection, cyber-security (AI4SEC), and more. This includes supervised machine learning approaches to provide, for example, predictive maintenance solutions that leverage data from various sensors and maintenance records to predict, for example, equipment failures, and unsupervised approaches based on anomaly detection in time-series and network data. Read more
Environmental Analytics
Environmental Analytics leverages data and analytical methods to gain a deeper understanding of environmental issues and help to achieve the sustainability goals as part of the European Green Deal. The scientific focus lies on developing innovative sensor fusion methods to integrate diverse types of sensors and modalities, and on combining model-driven and data-driven aspects by employing physics-informed machine learning approach to make predictions. Read more
Multimodal Information Retrieval
Multimodal Information Retrieval is dedicated to research methods and approaches to combine complementary multimodal information into task-specific knowledge representation, while bridging both the semantic and multimodal gap. This is achieved by exploring and combining task-independent fusion methods (e.g., knowledge graphs, named entity recognition, visual analysis) with appropriate methods for high-level knowledge modelling and indexing. Read more
