HIDA Lecture Series

HIDA Lectures @ HIDSS4Health: Synergies between mechanistic and data-driven models in medical research

Datum: 20.11.2024

Inhaltszusammenfassung:

Die Lecture behandelt die begrenzten Einsatzmöglichkeiten des maschinellen Lernens in der Medizin, obwohl in diesem Bereich erhebliche Fortschritte erzielt wurden. Im Mittelpunkt steht das „Big Data, aber Small Data“-Paradoxon: Daten sind häufig aufgrund rechtlicher, ethischer oder technischer Hürden schwer zugänglich und mangelhaft standardisiert. Zudem fehlt es oft an großen, qualitativ hochwertigen Datensätzen, während verzerrte Daten Fragen zur Fairness und Vollständigkeit der Algorithmen aufwerfen.

Die Lecture zeigt, wie mechanistische Modellierung und Simulation genutzt werden können, um diese Herausforderungen zu bewältigen, indem sie maschinelles Lernen integrieren. Anhand eines mechanistischen Modells der Herz-Elektrophysiologie wird demonstriert, wie synthetische EKG-Daten generiert und zur Optimierung von ML-Modellen eingesetzt werden können.

HIDA Lectures @ HIDSS4Health: Towards AI-Based Precision Neurology

Datum: 08.11.2024

Inhaltszusammenfassung: 

Neurodegenerative Erkrankungen (NDDs) nehmen weltweit zu, insbesondere in Entwicklungs- und Schwellenländern. Ihre Mechanismen sind bislang wenig erforscht, und erste pathologische Veränderungen gehen den Symptomen oft um Jahre voraus, was eine präzise Diagnose ohne spezialisierte Einrichtungen erschwert.

Der Vortrag beleuchtet, wie datengestützte Assistenzsysteme die Früherkennung und Behandlung solcher Krankheiten verbessern könnten. Anhand von Beispielen wird die Entwicklung eines KI-gestützten digitalen Zwillings vorgestellt, gefolgt von einer Diskussion aktueller Herausforderungen und Zukunftsperspektiven, insbesondere im Kontext der neu geförderten EU-Graduiertenschule „AI for Parkinson's Disease (AIPD)“.

HIDA Lectures @ HIDSS4Health: FAIR clinical IT-infrastructure for the virtual human twin

Datum: 30.10.2024

Inhaltszusammenfassung: Christian Niklas, klinischer Anästhesist und Postdoc am Institut für Medizinische Informatik des Universitätsklinikums Heidelberg, veranschaulicht in diesem Vortrag, wie die FAIR-Prinzipien zur Optimierung der klinischen IT-Infrastruktur beitragen und den Einsatz virtueller menschlicher Zwillinge (VHT) in der Medizin unterstützen können.

Er geht dabei auf die komplexen Herausforderungen der Datenintegration ein, beleuchtet Aspekte der semantischen und syntaktischen Interoperabilität und zeigt die Bedeutung einer nachhaltigen IT-Infrastruktur für die Förderung der personalisierten Medizin auf.

HIDA Lecture @ HEIBRIDS: "The future of governing equations"

Datum: 13.12.2023

Inhaltszusammenfassung: A major challenge in the study of dynamical systems is that of model discovery: turning data into reduced order models that are not just predictive, but provide insight into the nature of the underlying dynamical system that generated the data.

We introduce a number of data-driven strategies for discovering nonlinear multiscale dynamical systems and their embeddings from data. We consider two canonical cases: (i) systems for which we have full measurements of the governing variables, and (ii) systems for which we have incomplete measurements. For systems with full state measurements, we show that the recent sparse identification of nonlinear dynamical systems (SINDy) method can discover governing equations with relatively little data and introduce a sampling method that allows SINDy to scale efficiently to problems with multiple time scales, noise and parametric dependencies. For systems with incomplete observations, we show that the Hankel alternative view of Koopman (HAVOK) method, based on time-delay embedding coordinates and the dynamic mode decomposition, can be used to obtain a linear models and Koopman invariant measurement systems that nearly perfectly captures the dynamics of nonlinear quasiperiodic systems.

Neural networks are used in targeted ways to aid in the model reduction process. Together, these approaches provide a suite of mathematical strategies for reducing the data required to discover and model nonlinear multiscale systems.

HIDA Lecture @ HDS-LEE: "On Learning-Aware Mechanism Design"

Datum: 01.03.2023

Inhaltszusammenfassung: Statistical decisions are often given meaning in the context of other decisions, particularly when there are scarce resources to be shared.  Managing such sharing is one of the classical goals of microeconomics, and it is given new relevance in the modern setting of large, human-focused datasets, and in data-analytic contexts such as classifiers and recommendation systems. Michael I. Jordan will discuss several recent projects that aim to explore the interface between machine learning and microeconomics, including leader/follower dynamics in strategic classification, a Lyapunov theory for matching markets with transfers, and the use of contract theory as a way to design mechanisms that perform statistical inference.

HIDA Lecture @ MarDATA: "The Era of High-Resolution, Integrated Multi-Omics"

Datum: 10.11.2022

Abstract: The vast majority of life on our planet is microbial. An astonishing number of microbial organisms living in terrestrial and marine habitats represent a biomass that exceeds every living organism that can be seen by naked eye. They complement their numbers with their inconceivable diversity that allow microbes to synthesize or break down a wide array of chemical substrates, and govern biogeochemical cycles that make Earth a habitable planet for much less talented organisms (such as ourselves). While advances in molecular biology and sequencing technologies now enable unprecedented descriptions of the diversity of our naturally occurring microbial overlords, we are far from being able to make sense of the extremely complex datasets our environmental surveys yield almost on a daily basis. The purpose of this talk is to highlight some of the recent progress and challenges that are particularly suitable for data scientists to ponder.

HIDA Lecture @ SummerAcademy: "Data-Driven Inertial Sensing"

Datum: 16.09.2022

Abstract: The purpose of navigation is to determine the position, velocity, and orientation of manned and autonomous platforms, humans, and animals. Obtaining accurate navigation commonly requires fusion between several sensors, such as inertial sensors and global navigation satellite systems, in a model-based nonlinear estimation framework. Recently, data-driven approaches applied in various fields show state-of-the-art performance, compared to model-based methods.  In this talk, we address data-driven based navigation algorithms, recently derived at the autonomous navigation and sensor fusion lab. The purpose of those algorithms is to enhance common navigation and estimation tasks and open new possibilities for accurate and robust navigation. Data driven inertial navigation topics included in this talk will highlight hybrid learning and end to end learning approaches for different platforms and applications such as: pedestrian dead reckoning with inertial sensors, quadrotor dead recooking, learning vehicle trajectory uncertainty by hybrid models, and autonomous underwater vehicle navigation.

HIDA Lecture @ HEIBRiDS: "The Data Donation Project - How wearable sensors can help in dealing with the COVID-19 crisis"

Datum: 27.04.2022

Abstract: A theoretical physicist by training, Dirk Brockmann research focuses on complex systems at the interface of physics, life sciences and social sciences. He is particularly interested the application of dynamicals systems theory, stochastic processes and network science to infectious disease dynamics and related contagion processes. In this context he is currently investigating dynamical processes on time-dependent networks, complex contagion processes and the emergence of cooperation in evolutionary processes. He is known for his work on human mobility and its role on the global spread of infectious diseases. He talks here about the opportunities of infection disease research like the COVID-19 pandemic in using data from voluntary data donors.

HIDA Lecture @ MarDATA: "Machine-Learning-Model-Data-Integration for a Better Understanding of the Earth System"

Datum: 24.03.2021, 15:00 Uhr

Abstract: The Earth is a complex dynamic networked system. Machine learning, i.e. derivation of computational models from data, has already made important contributions to predict and understand components of the Earth system, specifically in climate, remote sensing and environmental sciences. For instance, classifications of land cover types, prediction of land-atmosphere and ocean-atmosphere exchange, or detection of extreme events have greatly benefited from these approaches. Such data-driven information has already changed how Earth system models are evaluated and further developed. However, many studies have not yet sufficiently addressed and exploited dynamic aspects of systems, such as memory effects for prediction and effects of spatial context, e.g. for classification and change detection. In particular new developments in deep learning offer great potential to overcome these limitations.

Yet, a key challenge and opportunity is to integrate (physical-biological) system modeling approaches with machine learning into hybrid modeling approaches, which combines physical consistency and machine learning versatility. A couple of examples are given with focus on the terrestrial biosphere, where the combination of system-based and machine-learning-based modelling helps our understanding of aspects of the Earth system.

HIDA Lectures @ HIDSS4Health: "AI-enabled Imaging"

Datum: 23.02.2022

Abstract: Artificial intelligence, in particular from the class of machine / deep learning, has shown great promise for applications in medical imaging. However, the success of AI-based techniques is limited by the availability and quality of the training data. A common approach is to train methods on well annotated and curated databases of high-quality image acquisitions, which then may fail on real patient cases in a hospital setting. Another problematic is the lack of sufficient numbers of clinical label annotations in the training data, or example for early markers of disease. In this talk I will present some of our recent approaches in cardiac magnetic resonance imaging (CMR) that aim to address some of these challenges, by using AI as an enabling technique for improved CMR image reconstruction, realistic CMR augmentation and further downstream tasks.

HIDA Lectures @ HIDSS4Health: "Data Science for Supporting Molecular Tumor Boards"

Datum: 26.01.22

Abstract: Molecular Tumor Boards are interdisciplinary teams of medical experts which come together to obtain therapy recommendations for complex oncological cases. In recent years, these recommendations are more and more based on molecular profiling of tumors, in particular the sequencing of cancer genes followed by the detection of genetic variants. The PREDICT project set out to develop systems and algorithms for supporting such decision-making. We will discuss the main results of this project i.e. (1) VIST, a search engine specifically constructed to find clinically relevant documents given a patient’s variant profile, (2) I-VIS, a framework for building integrated precision oncology databases, and (3) BRONCO, the first open and annotated corpus of German medical texts.  Building these systems required advanced NLP technologies, the application of modern machine learning methods, and robust data integration technologies.

HIDA Lectures @ HDS-LEE: "Machine Learning for Scientific Discovery, with Examples in Fluid Mechanics"

Termin: 07.12.2021, 17 Uhr

Abstract: This work describes how machine learning may be used to develop accurate and efficient nonlinear dynamical systems models for complex natural and engineered systems.  We explore the sparse identification of nonlinear dynamics (SINDy) algorithm, which identifies a minimal dynamical system model that balances model complexity with accuracy, avoiding overfitting.  This approach tends to promote models that are interpretable and generalizable, capturing the essential “physics” of the system.  We also discuss the importance of learning effective coordinate systems in which the dynamics may be expected to be sparse.  This sparse modeling approach will be demonstrated on a range of challenging modeling problems in fluid dynamics, and we will discuss how to incorporate these models into existing model-based control efforts.  Because fluid dynamics is central to transportation, health, and defense systems, we will emphasize the importance of machine learning solutions that are interpretable, explainable, generalizable, and that respect known physics.

HIDA Lecture @ HEIBRiDS: "TARS: Few-Shot Learning for Natural Language Processing"

Termin: 16. Juni 2021, 16 Uhr

Abstract: Machine learning models for natural language processing (NLP) are typically trained with very large amounts of labeled training data. However, such data is often not readily available and very expensive to produce. In this talk I present TARS, a novel approach in the research area of "few-shot learning" which allows us to train text classification models with little training data - or even none at all! I show how the proposed approach can be applied to a continual learning setup in which a single model learns a number of different tasks in sequence, with the goal of learning all tasks. Finally, I give a brief overview of the Flair NLP framework (https://github.com/flairNLP/flair) we develop in my group together with the open source community, and show how you can use TARS (and other NLP components in Flair) in your own research or industry projects.

HIDA Lecture @ DASHH: "Quantum Machine Learning in Chemical Compound Space"

Termin: 15. April 2021, 14 Uhr

Abstract: Many of the most relevant observables of matter depend explicitly on atomistic and electronic details, rendering a first principles approach to computational materials design mandatory. Alas, even when using high-performance computers, brute force high-throughput screening of material candidates is beyond any capacity for all but the simplest systems and properties due to the combinatorial nature of compound space, i.e. all the possible combinations of compositional and structural degrees of freedom. Consequently, efficient exploration algorithms exploit implicit redundancies and correlations. I will discuss recently developed statistical learning based approaches for interpolating quantum mechanical observables throughout compound space. Numerical results indicate promising performance in terms of efficiency, accuracy, scalability and transferability.