Projects

Dr. Georgia Koppe

Durstewitz D. Deutsches Zentrum für Luft- und Raumfahrt 01GQ2509A: NAILit – Neuro-inspirierte Künstliche Intelligenz für Lernen und Inferenz in nicht-stationären Umgebungen – Teilprojekt Entwicklung der KI Architektur". 12/2025-11/2028.

Es werden modernste KI‐Werkzeuge zur Rekonstruktion dynamischer Systeme (DSR) eingesetzt, um aus neurophysiologischen und verhaltensbezogenen Daten generative Surrogatmodelle abzuleiten, die eine detaillierte Analyse der computationalen und adaptiven Prozesse ermöglichen, die dem Online‐ (“In‐Context‐/Few‐Shot‐”) und kontinuierlichen Lernen zugrunde liegen. Im Fokus stehen fünf Verhaltensaufgaben, deren Aufbau eng an Benchmarks angelehnt ist, die zur Testung von KI‐Algorithmen verwendet werden, und die wechselnde Aufgabenregeln, Belohnungswahrscheinlichkeiten oder Stimulusbedingungen beinhalten. Die abgeleiteten dynamischen und Plastizitätsmechanismen werden in Deep‐Learning‐Architekturen implementiert, auf gängigen nicht‐stationären Benchmark‐Aufgaben getestet und mit anderen „state‐of‐the‐art“ KI‐Algorithmen verglichen. Darüber hinaus wird die Übertragung der aus Daten hergeleiteten Algorithmen in Energie‐ und Kodierungs‐effiziente spikende neuronale Netzwerke untersucht, sowie deren Anwendung in psychiatrischen Kontexten, in denen es um die Vorhersage von nicht‐stationären Krankheitsverläufen und die Steuerung von Neurofeedback‐Designs geht.

Reininghaus U. MWK - Ministerium für Wissenschaft Forschung und Kunst Baden-Württemberg : Reallabor Künstliche Intelligenz für digitale personalisierte psychische Gesundheitsförderung bei jungen Menschen. 01/2021-12/2023.

Spanagel R. DFG - Deutsche Forschungsgemeinschaft SFB/Transregio 265: Teilprojekt B08: Aversion discounting in animal models and human addiction. 07/2019-06/2023.

Durstewitz D, Koppe G. DFG - Deutsche Forschungsgemeinschaft : CRC TRR 265: Project A06: AI-based predictive neuro-behavioral modeling of individual trajectories in addiction. 07/2019-06/2023.

Abstract

Our project aims at inferring subject-level recurrent neural network (RNN) models of behavioral dynamics from multi-modal mobile data. We will develop methods that integrate data obtained from different modalities (such as EMA, accelerometer etc.) and follow distinct probability distributions and sampling rates. This framework will be applied to data from A01-A04 to identify and predict dynamical transitions between regaining & losing control, predict long-term trajectories, and identify subgroups based on inferred RNN parameters. Lastly, we aim at identifying crucial factors and drivers behind these transitions, as well as mechanisms governing the behavioral dynamics.

Kirsch P, Koppe G, Sommer WH. DFG - Deutsche Forschungsgemeinschaft : CRC TRR 265: Project B08: Aversion discounting in behavioral control in animal models and human addiction. 07/2019-06/2023.

Abstract

To date, reward discounting but not aversion discounting was examined in SUD. Our working hypothesis of increased temporal aversion discounting in AUD patients will be tested by novel tasks for reliable and quantitative assessment of aversion discounting in humans and animal models. We will study the underlying neurobiology of aversion discounting by fMRI in humans and calcium imaging microendoscopy in rats. Computational analyses will be used to model the decision-making processes and deliver a detailed and formal parametrization of aversion discounting on multiple levels of analysis. In the future, such information can be used for the development of therapeutic approaches that strengthen self-regulation and cognitive control

Reininghaus U. EU - Europäische Union 945263: IMMERSE - The implementation of Digital Mobile Mental Health in clinical care pathways: Towards person-centered care in psychiatry.

Abstract

The overarching aim of IMMERSE (Implementing Mobile MEntal health Recording Strategy for Europe) is to advance the transformation of mental health care in Europe into true person-centered care, focused on the needs of each individual seeking help for mental health problems, while giving them an active role in their treatment process and decision-making. In order to do so, IMMERSE has identified the Experience Sampling Methodology (ESM), a structured diary technique, as the methodology that puts the service user at the heart of their treatment. IMMERSE will integrate 20 years of research evidence on ESM into an innovative, clinical digital health tool, Digital Mobile Mental Health (DMMH), in close collaboration with stakeholders and extending it with mobile sensing data and innovative machine learning models. DMMH consists of an ESM app, assessing self-reports of mental state in daily life (ecological momentary assessment, EMA), a data-platform that allows the analysis of these data, and dashboard for visualization and feedback. IMMERSE will thoroughly evaluate strategies, processes and outcomes of DMMH implementation in a cluster Randomized Controlled Trial (cRCT) at 8 sites in 4 countries in Europe representing different contexts for implementation evaluation. At the same time, IMMERSE will identify and overcome key barriers and strengthen facilitators for implementation, transfer and scale-up of DMMH to routine mental health clinical practice by closely collaborating with relevant stakeholders, aligning the innovative DMMH tool to their needs. Similarly, the diverse ethical, legal and policy challenges and requirements will be identified and DMMH will be developed and implemented accordingly. Finally, IMMERSE is set out to do a cost-benefit analysis of the implementation and present a framework for future implementation of DMMH, including forecasting scenarios, aiming at a further scale-up of DMMH across 4 countries in Europe and beyond. IMMERSE thus offers a unique potential to significantly innovate mental health care in Europe.

Zentralinstitut für Seelische Gesundheit (ZI) - https://www.zi-mannheim.de