Publications & Projects
Applied statistics, signal processing, and machine learning across engineering, medical research, and industry, with a strong focus on methodological rigor and reproducible analysis. Selected publications and representative projects highlighting my role and methodological contributions.
Methodological examples
Representative examples illustrating analysis workflows, methodological approaches and outcomes from applied projects.
Bayesian estimation of discrete system states from high-dimensional data
Discrete system states are estimated from high-dimensional measurements using a probabilistic modeling framework. Dimensionality reduction via principal component analysis is combined with Gaussian mixture model–based likelihoods and sequential inference using a Hidden Markov Model.
Context: Academic research / Applied statistical signal processing
Role: Probabilistic model development, Bayesian inference design, and validation
Methods: Machine Learning · Sequential Bayesian inference · Discrete state estimation · Uncertainty quantification
Problem
In many technical and scientific systems, the underlying system state is discrete but only indirectly observable through high-dimensional measurement data. The key challenge lies in reliably inferring these latent states while accounting for noise, uncertainty, and high dimensionality.
Approach
- Dimensionality reduction of high-dimensional measurement data using principal component analysis.
- Gaussian mixture model–based likelihood linking latent discrete states to reduced observation space.
- Sequential Bayesian inference using a Hidden Markov Model with explicit uncertainty quantification.
- Validation of state estimation performance using simulated and experimental data.
Figures
Figure 1: Probabilistic inference pipeline for estimating discrete system states from high-dimensional measurements. Principal component analysis is used for dimensionality reduction, Gaussian mixture models define state-dependent likelihoods, and a Hidden Markov Model enables sequential Bayesian inference to obtain posterior state probabilities.
Figure 2: Example of dimensionally reduced measurement data and corresponding posterior state probabilities over time. The upper plot shows the first principal component of the reduced measurement vector, while the lower plot depicts the time-resolved posterior probabilities of the discrete system states obtained from sequential Bayesian inference.
Figure 1: Probabilistic inference pipeline for estimating discrete system states from high-dimensional measurements. Principal component analysis is used for dimensionality reduction, Gaussian mixture models define state-dependent likelihoods, and a Hidden Markov Model enables sequential Bayesian inference to obtain posterior state probabilities.
Figure 2: Example of dimensionally reduced measurement data and corresponding posterior state probabilities over time. The upper plot shows the first principal component of the reduced measurement vector, while the lower plot depicts the time-resolved posterior probabilities of the discrete system states obtained from sequential Bayesian inference.
Outcome
The proposed framework enables robust state classification while providing posterior state probabilities rather than point estimates, allowing uncertainty-aware downstream decision making.
Selected Projects
An overview of applied projects from research and industry, highlighting methodological work in statistical analysis, signal processing, machine learning, and scientific software development.
Bayesian estimation of discrete system states from high-dimensional data
Discrete system states are estimated from high-dimensional measurements using a probabilistic modeling framework. Principal component analysis is combined with Gaussian mixture model–based likelihoods and sequential inference via a Hidden Markov Model.
Role: Probabilistic model development, algorithm design, and implementation, experimental evaluation, data analysis, and validation
Methods: Machine Learning · Sequential Bayesian inference · Discrete state estimation · Uncertainty quantification
Nonlinear Trend Filtering for Noisy Signals with Sharp Transients
A nonlinear trend filtering approach is used to extract smooth trends from noisy signals containing sharp transients and slowly varying segments. Edge-preserving regularization enables robust noise suppression while retaining relevant signal dynamics.
Role: Method development, algorithm design, implementation, experimental evaluation, data analysis, and validation
Methods: Signal processing · Nonlinear Filtering · Robust Trend Estimation · ℓ₁ Regularization
Neural Network–Based Acceleration of 3D Physics Simulations
A neural network surrogate model is developed to approximate a three-dimensional finite element–based physical simulation for real-time use. Leveraging a validated physics model for extensive training and testing results in a robust surrogate that significantly accelerates simulation and subsequent model-based inference.
Role: Surrogate model design, neural network development, training, validation, and integration
Methods: Machine Learning · Surrogate Modeling · Neural Networks · Model Acceleration
Bayesian estimation of discrete system states from high-dimensional data
Discrete system states are estimated from high-dimensional measurements using a probabilistic modeling framework. Principal component analysis is combined with Gaussian mixture model–based likelihoods and sequential inference via a Hidden Markov Model.
Role: Probabilistic model development, algorithm design, and implementation, experimental evaluation, data analysis, and validation
Methods: Machine Learning · Sequential Bayesian inference · Discrete state estimation · Uncertainty quantification
Nonlinear Trend Filtering for Noisy Signals with Sharp Transients
A nonlinear trend filtering approach is used to extract smooth trends from noisy signals containing sharp transients and slowly varying segments. Edge-preserving regularization enables robust noise suppression while retaining relevant signal dynamics.
Role: Method development, algorithm design, implementation, experimental evaluation, data analysis, and validation
Methods: Signal processing · Nonlinear Filtering · Robust Trend Estimation · ℓ₁ Regularization
Neural Network–Based Acceleration of 3D Physics Simulations
A neural network surrogate model is developed to approximate a three-dimensional finite element–based physical simulation for real-time use. Leveraging a validated physics model for extensive training and testing results in a robust surrogate that significantly accelerates simulation and subsequent model-based inference.
Role: Surrogate model design, neural network development, training, validation, and integration
Methods: Machine Learning · Surrogate Modeling · Neural Networks · Model Acceleration
Selected Publications
Selected peer-reviewed publications in engineering, measurement science, and medical research with a focus on applied statistical and signal-based methodologies. Several of these projects involved methodological contributions during study design, statistical reporting, and manuscript preparation.
Learn more about scientific support →
Medical & Biostatistics
S. Kurath-Koller, D. Scherr, N. Oeffl et al., „Feasibility and diagnostic agreement of Apple Watch and KardiaMobile electrocardiograms compared with standard 12-lead ECG in children,“ in Scientific Reports, 2025
Contribution: Statistical methodology
DOI · Publisher · Google Scholar
M. Bruckner, T. Suppan, E. Suppan et al., „Brain oxygenation monitoring during neonatal stabilization and resuscitation and its potential for improving preterm infant outcomes: a systematic review and meta-analysis with Bayesian analysis,“ in European Journal of Pediatrics, vol 184 (305), 2025
Contribution: Statistical methodology, model development, contribution to the methods section
DOI · Publisher · Google Scholar
Engineering & Measurement Science
M. Neumayer, T. Suppan, T. Bretterklieber et al., „Fast numerical techniques for FE simulations in electrical capacitance tomography ,“ in COMPEL, vol. 42(5), 2023
Contribution: Support in method development and implementation.
DOI · Publisher · Google Scholar
T. Suppan, M. Neumayer, T. Bretterklieber et al., „Electrical capacitance tomography-based estimation of slug flow parameters in horizontally aligned pneumatic conveyors,“ in Powder Technology, vol. 420, 2023
Contribution: Method development, experimental measurements, data analysis, and manuscript preparation
DOI · Publisher · Google Scholar
T. Suppan, M. Neumayer, T. Bretterklieber et al., „Thermal Drifts of Capacitive Flow Meters: Analysis of Effects and Model-Based Compensation,“ in IEEE Transactions on Instrumentation and Measurement, vol. 70, 2021
Contribution: Method development, experimental measurements, data analysis, and manuscript preparation
DOI · Publisher · Google Scholar
M. Neumayer, T. Suppan, T. Bretterklieber et al., „Statistical solution of inverse problems using a state reduction ,“ in COMPEL, vol. 38(5), 2019
Contribution: Support in method development and implementation.
DOI · Publisher · Google Scholar
T. Suppan, M. Neumayer, T. Bretterklieber et al., „Prior design for tomographic volume fraction estimation in pneumatic conveying systems from capacitive data,“ in Transactions of the Institute of Measurement and Control, vol. (42)4, 2019
Contribution: Method development, experimental measurements, data analysis, and manuscript preparation
DOI · Publisher · Google Scholar
View full publication record on Google Scholar or ORCID
Discuss Future Collaboration
If you are working on a scientific or technical problem that requires rigorous analysis and clear documentation, I’m happy to exchange ideas and discuss potential methodological approaches.
Free initial call · NDA possible · Remote collaboration across AT/DE/EU