About

Felix J. Herrmann graduated from Delft University of Technology in 1992 and received his Ph.D. in engineering physics from that same institution in 1997. After research positions at Stanford University and the Massachusetts Institute of Technology, he became faculty at the University of British Columbia in 2002. In 2017, he joined the Georgia Institute of Technology, where he is now a Georgia research Alliance Scholar Chair in Energy, cross-appointed between the Schools of Earth & Atmospheric Sciences, Computational Science & Engineering, and Electrical & Computer Engineering. His cross-disciplinary research program spans several areas of computational imaging, including seismic and medical imaging. Dr. Herrmann is widely known for tackling challenging problems in the imaging sciences by adapting techniques from randomized linear algebra, PDE-constrained and convex optimization, high-performance computing, machine learning, and uncertainty quantification. Over his career, he has been responsible for several cost-saving innovations in industrial time-lapse seismic data acquisition and wave-equation based imaging. In 2019, he toured the world presenting the SEG Distinguished Lecture 'Sometimes it pays to be cheap – Compressive time-lapse seismic data acquisition'. In 2020, he received the SEG Reginald Fessenden Award for his contributions to seismic data acquisition with compressive sensing. At Georgia Tech, he leads the Seismic Laboratory for Imaging and modeling and is co-founder and director of the Center for Machine Learning for Seismic (ML4Seismic), which aims to foster industrial research partnerships to drive innovations in artificial intelligence-assisted seismic imaging, interpretation, analysis, and time-lapse monitoring.

Research topics

• Computer science
• Algorithm
• Geology
• Artificial intelligence
• Mathematical optimization

Selected publications

• Sensitivity-aware rock physics enhanced digital shadow for underground-energy storage monitoring

  2026-02-10

  articleSenior author

  Underground energy storage, which includes storage of hydrogen, compressed air, and CO2, requires careful monitoring to track potential leakage pathways, a situation where time-lapse seismic imaging alone may be inadequate. A recently developed Digital Shadow (DS) enhances forecasting using machine learning and Bayesian inference, yet their accuracy depends on assumed rock physics models, the mismatch of which can lead to unreliable predictions for the reservoir’s state (saturation/pressure). Augmenting DS training with multiple rock physics models mitigates errors but averages over uncertainties, obscuring their sources. To address this challenge, we introduce context-aware sensitivity analysis inspired by amortized Bayesian inference, allowing the DS to learn explicit dependencies between seismic data, the reservoir state, e.g., CO2 saturation, and rock physics models. At inference time, this approach allows for real-time “what if” scenario testing rather than relying on costly retraining, thereby enhancing interpretability and decision-making for safer, more reliable underground storage.

  PublisherDOI

• Data‐Driven Feedback Identifies Focused Ultrasound Exposure Regimens for Improved Nanotheranostic Targeting of the Brain

  Advanced Science · 2026-01-07 · 1 citations

  articleOpen access

  The blood-brain barrier (BBB) renders the delivery of nanomedicine in the brain ineffective and the detection of circulating disease-related DNA from the brain unreliable. Here, we demonstrate that microbubble-enhanced focused ultrasound (MB-FUS) mediated BBB opening, supported by large-data models predict sonication regimens for safe and effective BBB opening. Importantly, a closed-loop MB-FUS controller augmented by machine learning (ML-CL) expands the treatment window, as compared to conventional controllers, by persistently and proactively maximizing the BBB permeability while preventing tissue damage. By successfully scaling up from mice to rats and from healthy to diseased brains (glioma), ML-CL rendered the BBB permeable to large nanoparticles and markedly improved the release and detection of reporter gene DNA from tumors in blood. Together, our findings reveal the potential of data-driven feedback to support the development of next-generation AI-powered ultrasound systems for safe, robust, and efficient nanotheranostic targeting and treatment of brain diseases.

  PublisherOA PDFDOI

• Well2Flow: Reconstruction of reservoir states from sparse wells using score-based generative models

  2026-02-10

  articleSenior author

  This study investigates the use of score-based generative models for reservoir simulation, with a focus on reconstructing spatially varying permeability and saturation fields in saline aquifers, inferred from sparse observations at two well locations. By modeling the joint distribution of permeability and saturation derived from high-fidelity reservoir simulations, the proposed neural network is trained to learn the complex spatiotemporal dynamics governing multiphase fluid flow in porous media. During inference, the framework effectively reconstructs both permeability and saturation fields by conditioning on sparse vertical profiles extracted from well log data. This approach introduces a novel methodology for incorporating physical constraints and well log guidance into generative models, significantly enhancing the accuracy and physical plausibility of the reconstructed subsurface states. Furthermore, the framework demonstrates strong generalization capabilities across varying geological scenarios, highlighting its potential for practical deployment in data-scarce reservoir management tasks.

  PublisherDOI

• Efficient and scalable posterior surrogate for seismic inversion via wavelet score-based generative models

  2026-02-10

  articleSenior author

  Seismic inversion poses significant computational challenges due to its high dimensionality and non-unique solutions. We propose a novel method integrating the Wavelet Score-Based Generative Model (WSGM) with Simulation-Based Inference (SBI) to enable efficient posterior sampling for full-waveform inference. Our approach reduces memory requirements (≍ 50%) and significantly decreases sampling time (≍ 73%) compared to standard score-based diffusion models, while preserving accuracy. Furthermore, WSGM naturally supports the generation of velocity models at multiple resolutions, leveraging its hierarchical structure. Experimental results on pairs of synthetic seismic images and velocity models demonstrate that our method enables posterior sampling for large-scale 2D geophysica problems and facilitates the assessment of uncertainties relevant to subsurface characterization.

  PublisherDOI

• Power-scaled Bayesian inference with score-based generative models

  2026-02-10

  articleOpen accessSenior author

  We propose a score-based generative algorithm for sampling from power-scaled priors and likelihoods within the Bayesian inference framework. Our algorithm enables flexible control over prior–likelihood influence without requiring retraining for different power scaling configurations. Specifically, we focus on generative models to synthesize seismic velocity models conditioned on imaged seismic. Our method enables sensitivity analysis by sampling from intermediate power posteriors, allowing us to assess the relative influence of the prior and likelihood on samples of the posterior distribution. Through a comprehensive set of experiments, we evaluate the effects of varying the power parameter in different settings: applying it solely to the prior, to the likelihood of a Bayesian formulation, and to both simultaneously. The results show that increasing the power of the likelihood up to a certain threshold improves the fidelity of posterior samples to the conditioning data (e.g., RTM), while decreasing the prior power promotes greater structural diversity among samples. Moreover, we find that moderate scaling of the likelihood leads to a reduced shot data residual, confirming its utility in posterior refinement.

  PublisherOA PDFDOI

• Full-waveform variational inference with full common-image gathers and diffusion network

  2026-02-10

  articleSenior author

  Accurate seismic imaging and velocity estimation are essential for subsurface characterization. Conventional inversion techniques, such as full-waveform inversion, remain computationally expensive and sensitive to initial velocity models. To address these challenges, we propose a simulation-based inference framework with conditional elucidated diffusion models for posterior velocity-model sampling. Our approach incorporates both horizontal and vertical subsurface offset commonimage gathers to capture a broader range of reflector geometries, including gently dipping structures and steep dipping layers. Additionally, we introduce the background-velocity model as an input condition to enhance generalization across varying geological settings. We evaluate our method on the SEAM dataset, which features complex salt geometries, using a patchbased training approach. Experimental results demonstrate that adding the background-velocity model as an additional conditioning variable significantly enhances performance, improving SSIM from 0. 717 to 0. 733 and reducing RMSE from 0. 381km/s to 0. 274km/s. Furthermore, uncertainty quantification analysis shows that our proposed approach yields bettercalibrated uncertainty estimates, reducing uncertainty calibration error from 6. 68km/s to 3. 91km/s. These results show robust amortized seismic inversion with uncertainty quantification.

  PublisherDOI

• Probabilistic Joint Recovery Method for CO$_2$ Plume Monitoring

  ArXiv.org · 2025-01-30

  preprintOpen accessSenior author

  Reducing CO$_2$ emissions is crucial to mitigating climate change. Carbon Capture and Storage (CCS) is one of the few technologies capable of achieving net-negative CO$_2$ emissions. However, predicting fluid flow patterns in CCS remains challenging due to uncertainties in CO$_2$ plume dynamics and reservoir properties. Building on existing seismic imaging methods like the Joint Recovery Method (JRM), which lacks uncertainty quantification, we propose the Probabilistic Joint Recovery Method (pJRM). By estimating posterior distributions across surveys using a shared generative model, pJRM provides uncertainty information to improve risk assessment in CCS projects.

  PublisherOA PDFDOI

• A reduced-order derivative-informed neural operator for subsurface fluid-flow

  ArXiv.org · 2025-09-17

  preprintOpen accessSenior author

  Neural operators have emerged as cost-effective surrogates for expensive fluid-flow simulators, particularly in computationally intensive tasks such as permeability inversion from time-lapse seismic data, and uncertainty quantification. In these applications, the fidelity of the surrogate's gradients with respect to system parameters is crucial, as the accuracy of downstream tasks, such as optimization and Bayesian inference, relies directly on the quality of the derivative information. Recent advances in physics-informed methods have leveraged derivative information to improve surrogate accuracy. However, incorporating explicit Jacobians can become computationally prohibitive, as the complexity typically scales quadratically with the number of input parameters. To address this limitation, we propose DeFINO (Derivative-based Fisher-score Informed Neural Operator), a reduced-order, derivative-informed training framework. DeFINO integrates Fourier neural operators (FNOs) with a novel derivative-based training strategy guided by the Fisher Information Matrix (FIM). By projecting Jacobians onto dominant eigen-directions identified by the FIM, DeFINO captures critical sensitivity information directly informed by observational data, significantly reducing computational expense. We validate DeFINO through synthetic experiments in the context of subsurface multi-phase fluid-flow, demonstrating improvements in gradient accuracy while maintaining robust forward predictions of underlying fluid dynamics. These results highlight DeFINO's potential to offer practical, scalable solutions for inversion problems in complex real-world scenarios, all at substantially reduced computational cost.

  PublisherOA PDFDOI

• Data-driven feedback augments ultrasound nanotheranostics in brain tumors

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-05-07

  preprintOpen access

  The blood-brain barrier (BBB) renders the delivery of nanomedicine in the brain ineffective and the detection of circulating disease-related DNA from the brain unreliable. Here, we show that the acoustic emission content of focused ultrasound-controlled microbubble dynamics (MB-FUS) incorporates precursor signals that allow large-data models to predict sonication regimens for safe and effective BBB opening. Crucially, closed-loop MB-FUS controller augmented by machine learning (ML-CL) expands the treatment window (4-fold), as compared to conventional controllers, by persistently and proactively maximizing the BBB permeability while preventing tissue damage. By successfully scaling up from mice to rats and from healthy to diseased brains (glioma), ML-CL rendered the BBB permeable to large nanoparticles and markedly improved the release and detection of tumor DNA in plasma. Together, our findings reveal the potential of data-driven feedback to support the development of next-generation AI-powered ultrasound systems for safe, robust, and efficient nanotheranostic targeting of brain diseases.

  PublisherOA PDFDOI

• ASPIRE: iterative amortized posterior inference for Bayesian inverse problems

  Inverse Problems · 2025-02-26 · 2 citations

  articleOpen accessSenior author

  Abstract Due to their uncertainty quantification, Bayesian solutions to inverse problems are the framework of choice in applications that are risk averse. These benefits come at the cost of computations that are in general, intractable. New advances in machine learning and variational inference (VI) have lowered this computational barrier by leveraging data-driven learning. Two VI paradigms have emerged that represent different tradeoffs: amortized and non-amortized. Amortized VI can produce fast results but due to generalizing to many observed datasets it produces suboptimal inference results. Non-amortized VI is slower at inference but finds better posterior approximations since it is specialized towards a single observed dataset. Current amortized VI techniques run into a sub-optimality wall that cannot be improved without more expressive neural networks or extra training data. We present a solution that enables iterative improvement of amortized posteriors that uses the same networks architectures and training data. The benefits of our method requires extra computations but these remain frugal since they are based on physics-hybrid methods and summary statistics. Importantly, these computations remain mostly offline thus our method maintains cheap and reusable online evaluation while bridging the optimality gap between these two paradigms. We denote our proposed method ASPIRE - A mortized posteriors with S ummaries that are P hysics-based and I teratively RE fined. We first validate our method on a stylized problem with a known posterior then demonstrate its practical use on a high-dimensional and nonlinear transcranial medical imaging problem with ultrasound. Compared with the baseline and previous methods in the literature, ASPIRE stands out as an computationally efficient and high-fidelity method for posterior inference.

  PublisherDOI

Frequent coauthors

• Mathias Louboutin

  84 shared

• Ali Siahkoohi

  58 shared

• Rajiv Kumar

  52 shared

• Tristan van Leeuwen

  Centrum Wiskunde & Informatica

  45 shared

• Philipp Witte

  44 shared

• Gabrio Rizzuti

  41 shared

• Gilles Hennenfent

  Chevron (United States)

  41 shared

• Deli Wang

  Jilin Province Science and Technology Department

  36 shared

Education

• DOCTOR OF ENGINEERING, Applied Physics

  Technische Universiteit Delft

  1997

Awards & honors

• SEG Reginald Fessenden Award (2020)
• SEG Distinguished Lecture "Sometimes it pays to be cheap – C…