Hasso-Plattner-Institut
Prof. Dr.-Ing. Bert Arnrich
  
 

AI-LAB-ITSE

AI laboratory for methodology, technology and teaching in IT systems technology for the analysis, planning and construction of AI-based complex IT systems

Funded by the Federal Ministry of Education and Research

Bjarne Pfitzner, Lin Zhou

Artificial intelligence (AI) is a key discipline that enables leap innovations in business and society in almost all areas of application. With the help of AI, applications and systems are enabled to cope with ever-increasing amounts of data (“Big Data”) and to analyze, interpret and evaluate them with great precision.

The AI ​​laboratory of the Hasso Plattner Institute (HPI) focuses on the planning, development and optimization of AI applications and AI systems - computer science and especially software engineering are facing major challenges and a fundamental change here.

The AI ​​lab at HPI deals, for example, with AI-suitable, resource and energy-efficient software architectures, important AI processes such as machine learning and deep learning, reasoning and problem solving, AI-specific data management and knowledge representation as well as AI-supported image and video evaluation. Another focus is on AI-relevant application areas such as digital health, personalized medicine, genomics and software, media and geodata analysis.

There are four work packages included in the AI lab:

  1. Federated Learning - Development and evaluation of ML processes for confidential data, which can only be collected centrally to a limited extent. A typical application scenario are clinical multicenter studies, which are often associated with a lengthy and often only limited central data collection from all partners involved. The aim is to develop methods for distributed learning in which learning takes place locally at the location of the data storage without the need to export clinical data and collect it centrally.
  2. Internet-of-Health-Things - Development and evaluation of AI methods that enable the reliable detection of health-relevant data from everyday life. The procedures are based on multimodal sensor data from everyday devices that are connected to the Internet independently or via a sensor hub (e.g. portable heart rate monitor, EKG chest strap, EEG headband, WLAN scales, etc.). The AI ​​methods to be developed must be able to adapt to the strongly fluctuating data quality of everyday data.
  3. Connected Health Patient Record - Fusion of electronic patient records with health-relevant everyday data. This should enable the design of AI models that can be used in particular for the prevention of chronic diseases, for the evaluation of therapies and for the early detection of health risks. AI models must be developed that can incorporate both data from existing electronic patient records and sensor data from everyday life.
  4. Transparent AI - Development of processes for an understandable representation of the operation of machine learning algorithms. This should enable in particular doctors and patients to better understand machine learning processes. The entire pipeline from preprocessing to modeling and inference should be included.