Victor Pankratius
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Victor Pankratius has a total of 32 co-authors. Here are the top co-authors by number of shared publications:
- Cody M. Rude
Department of Earth, Atmospheric, and Planetary Sciences Massachusetts Institute of Technology Cambridge MA USA
5 shared publications - Guillaume Rongier
Delft University of Technology, Unknown, Delft, Netherlands [NL
4 shared publications - Michael Gowanlock
Northern Arizona University
4 shared publications - Philip J. Erickson
Haystack Observatory, Massachusetts Institute of Technology, Westford, MA, USA
4 shared publications - Thomas A. Herring
MIT, Cambridge, MA, USA
4 shared publications - Christoph Wieland
Bosch Sensortec GmbH, Reutlingen, Germany
3 shared publications
Latest Publications
- Open Access
- 3 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (19) , 36843-36854
Inertial Signal Forecasting With Foundation Model Techniques
This paper introduces Dual-View FM, a generative pre-trained Foundation Model (FM) specifically designed for forecasting human inertial signals from body-worn sensors. Unlike traditional Transformer-based approaches, Dual-View FM employs a novel architecture that mixes global and local signal properties in the time and feature domains to build a comprehensive understanding of inertial activity data. The global view identifies general trends in human inertial signals, while the local view addresses their spontaneous nature by individually analyzing short, sensor-specific signal frames. Our evaluation on eight established datasets shows that Dual-View FM accurately forecasts human inertial signals from previously unseen test datasets that involve various sensors, activities, and users. The pre-trained models even outperform a production-ready FM for time series forecasting by up to 36 % in mean absolute error (MAE), demonstrating the advantages of domain-specific FMs. Fine-tuning further enhances Dual-View FM’s forecasting capabilities so that it surpasses other state-of-the-art architectures by more than 7 % in MAE.
- 3 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (19) , 36843-36854
Open AccessInertial Signal Forecasting With Foundation Model Techniques
This paper introduces Dual-View FM, a generative pre-trained Foundation Model (FM) specifically designed for forecasting human inertial signals from body-worn sensors. Unlike traditional Transformer-based approaches, Dual-View FM employs a novel architecture that mixes global and local signal properties in the time and feature domains to build a comprehensive understanding of inertial activity data. The global view identifies general trends in human inertial signals, while the local view addresses their spontaneous nature by individually analyzing short, sensor-specific signal frames. Our evaluation on eight established datasets shows that Dual-View FM accurately forecasts human inertial signals from previously unseen test datasets that involve various sensors, activities, and users. The pre-trained models even outperform a production-ready FM for time series forecasting by up to 36 % in mean absolute error (MAE), demonstrating the advantages of domain-specific FMs. Fine-tuning further enhances Dual-View FM’s forecasting capabilities so that it surpasses other state-of-the-art architectures by more than 7 % in MAE.
- Open Access
- 26 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (14) , 26920-26930
Enhancing Sensor-Based Human Activity Recognition With Multiresolution Wavelet-Attention
Sensor-based Human Activity Recognition (SHAR) is a research field that leverages wearable sensors for the continuous monitoring of human behavior, for example in sports, household, and healthcare environments. This paper introduces a novel SHAR approach that extends data-driven methods with sophisticated signal analysis. We propose a pre-trained Multi-Resolution Wavelet-Attention (MRWA) neural network block that learns multi-level Wavelet coefficients from SHAR signals and refines them using a cross-attention mechanism. The evaluation on eight widely-used benchmark datasets demonstrates that the combination of MRWA with existing SHAR classifiers significantly improves the detection of key signal features, boosting macro F1-scores by up to 0.286 points for DeepConvLSTM, 0.079 points for GRU, and 0.033 points for TinyHAR, all while increasing their inference times by less than 70 milliseconds. This keeps the MRWA-enriched classifiers efficient for deployment on current smartwatches. Additionally, MRWA improves the robustness and reliability of SHAR classifiers by keeping the standard deviations of the F1-scores low across multiple scenarios, thus providing a more accurate and efficient solution for seamless activity recognition.
- 26 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (14) , 26920-26930
Open AccessEnhancing Sensor-Based Human Activity Recognition With Multiresolution Wavelet-Attention
Sensor-based Human Activity Recognition (SHAR) is a research field that leverages wearable sensors for the continuous monitoring of human behavior, for example in sports, household, and healthcare environments. This paper introduces a novel SHAR approach that extends data-driven methods with sophisticated signal analysis. We propose a pre-trained Multi-Resolution Wavelet-Attention (MRWA) neural network block that learns multi-level Wavelet coefficients from SHAR signals and refines them using a cross-attention mechanism. The evaluation on eight widely-used benchmark datasets demonstrates that the combination of MRWA with existing SHAR classifiers significantly improves the detection of key signal features, boosting macro F1-scores by up to 0.286 points for DeepConvLSTM, 0.079 points for GRU, and 0.033 points for TinyHAR, all while increasing their inference times by less than 70 milliseconds. This keeps the MRWA-enriched classifiers efficient for deployment on current smartwatches. Additionally, MRWA improves the robustness and reliability of SHAR classifiers by keeping the standard deviations of the F1-scores low across multiple scenarios, thus providing a more accurate and efficient solution for seamless activity recognition.
- Open Access
Detection and Validation of Macro-Activities in Human Inertial Signals Using Graph Link Prediction
With the continuous development of new wearable devices, sensor-based human activity recognition is enjoying enormous popularity in research and industry. The signals from inertial sensors allow for the detection, classification, and analysis of human activities such as jogging, cycling, or swimming. However, human activity recognition is often limited to basic activities that occur in short, predetermined periods of time (sliding windows). Complex macro-activities, such as multi-step sports exercises or multi-step cooking recipes, are still only considered to a limited extent, while some works have investigated the classification of macro-activities, the automated understanding of how the underlying micro-activities interact remains an open challenge. This study addresses this gap through the application of graph link prediction, a well-known concept in graph theory and graph neural networks (GNNs). To this end, the presented approach transforms micro-activity sequences into micro-activity graphs that are then processed with a GNN. The evaluation on two derived real-world data sets shows that graph link prediction enables the accurate identification of interactions between micro-activities and the precise validation of composite macro-activities based on learned graph embeddings. Furthermore, this work shows that GNNs can benefit from positional encodings in sequence recognition tasks.
Open AccessDetection and Validation of Macro-Activities in Human Inertial Signals Using Graph Link Prediction
With the continuous development of new wearable devices, sensor-based human activity recognition is enjoying enormous popularity in research and industry. The signals from inertial sensors allow for the detection, classification, and analysis of human activities such as jogging, cycling, or swimming. However, human activity recognition is often limited to basic activities that occur in short, predetermined periods of time (sliding windows). Complex macro-activities, such as multi-step sports exercises or multi-step cooking recipes, are still only considered to a limited extent, while some works have investigated the classification of macro-activities, the automated understanding of how the underlying micro-activities interact remains an open challenge. This study addresses this gap through the application of graph link prediction, a well-known concept in graph theory and graph neural networks (GNNs). To this end, the presented approach transforms micro-activity sequences into micro-activity graphs that are then processed with a GNN. The evaluation on two derived real-world data sets shows that graph link prediction enables the accurate identification of interactions between micro-activities and the precise validation of composite macro-activities based on learned graph embeddings. Furthermore, this work shows that GNNs can benefit from positional encodings in sequence recognition tasks.
- Open Access
- 9 November 2020
- Published by California Digital Library (CDL) in EarthArXiv
An attempt at improving atmospheric corrections\\in InSAR using cycle-consistent adversarial networks
Interferometry from satellite radar has thrived as a major asset to study surface deformations from earthquakes, volcanoes, aquifers, glaciers, landslides, etc. Most signals recorded in an interferogram have precise enough models to remove them almost completely. Yet, current models still fail to capture the full range and scales of variations of atmospheric perturbations. This work explores the use of an image-to-image translation method, CycleGAN, to learn a function that wraps and improves an imperfect model for atmospheric correction. CycleGAN is a generative adversarial network, in which discriminators determine which images are real, and generators produce images to fool the discriminators. Training the discriminators and generators against each other improves the resulting translation function. We have tested this approach using Sentinel-1A data gathered around Puebla, Mexico, which was the starting point of the 2017 earthquake that devastated Mexico City. CycleGAN can generate visually compelling fringes, including small-scale perturbations that are absent from the atmospheric models, while limiting noisy areas. However, it fails to capture the variations of amplitude behind the fringes, especially at large scale, and the generated interferograms remain too different from the real interferograms. Solving that amplitude issue could create practical applications for a CycleGAN-type method in atmospheric correction or phase unwrapping. The large amount of InSAR data and the continuous progress of deep learning methods provide ample opportunity for improvement.
- 9 November 2020
- Published by California Digital Library (CDL) in EarthArXiv
Open AccessAn attempt at improving atmospheric corrections\\in InSAR using cycle-consistent adversarial networks
Interferometry from satellite radar has thrived as a major asset to study surface deformations from earthquakes, volcanoes, aquifers, glaciers, landslides, etc. Most signals recorded in an interferogram have precise enough models to remove them almost completely. Yet, current models still fail to capture the full range and scales of variations of atmospheric perturbations. This work explores the use of an image-to-image translation method, CycleGAN, to learn a function that wraps and improves an imperfect model for atmospheric correction. CycleGAN is a generative adversarial network, in which discriminators determine which images are real, and generators produce images to fool the discriminators. Training the discriminators and generators against each other improves the resulting translation function. We have tested this approach using Sentinel-1A data gathered around Puebla, Mexico, which was the starting point of the 2017 earthquake that devastated Mexico City. CycleGAN can generate visually compelling fringes, including small-scale perturbations that are absent from the atmospheric models, while limiting noisy areas. However, it fails to capture the variations of amplitude behind the fringes, especially at large scale, and the generated interferograms remain too different from the real interferograms. Solving that amplitude issue could create practical applications for a CycleGAN-type method in atmospheric correction or phase unwrapping. The large amount of InSAR data and the continuous progress of deep learning methods provide ample opportunity for improvement.
- Open Access
- 21 December 2019
- Published by American Geophysical Union (AGU) in Earth and Space Science
- Vol. 6 (12) , 2671-2683
Generative Modeling of InSAR Interferograms
Interferometric synthetic aperture radar (InSAR) has become an essential technique to detect surface variations due to volcanoes, earthquakes, landslides, glaciers, and aquifers. However, Earth's ionosphere, atmosphere, vegetation, surface runoff, etc., introduce noise that requires post‐processing to separate its components. This work defines a generator to create interferograms that include each of those components. Our approach leverages deformation models with real data, either directly or through machine learning using geostatistical methods. These methods result from previous developments to more efficiently and better simulate spatial variables and could replace some statistical approaches used in InSAR processing. We illustrate the use of the generator to simulate an artificial interferogram based on the 2015 Illapel earthquake and discuss the improved performance offered by geostatistical approaches compared with classical statistical ones. The generator establishes a tool for multiple applications (1) to evaluate InSAR correction workflows in controlled scenarios with known ground truth; (2) to develop training sets and generative methods for machine learning algorithms; and (3) to educate on InSAR and its principles.
- 21 December 2019
- Published by American Geophysical Union (AGU) in Earth and Space Science
- Vol. 6 (12) , 2671-2683
Open AccessGenerative Modeling of InSAR Interferograms
Interferometric synthetic aperture radar (InSAR) has become an essential technique to detect surface variations due to volcanoes, earthquakes, landslides, glaciers, and aquifers. However, Earth's ionosphere, atmosphere, vegetation, surface runoff, etc., introduce noise that requires post‐processing to separate its components. This work defines a generator to create interferograms that include each of those components. Our approach leverages deformation models with real data, either directly or through machine learning using geostatistical methods. These methods result from previous developments to more efficiently and better simulate spatial variables and could replace some statistical approaches used in InSAR processing. We illustrate the use of the generator to simulate an artificial interferogram based on the 2015 Illapel earthquake and discuss the improved performance offered by geostatistical approaches compared with classical statistical ones. The generator establishes a tool for multiple applications (1) to evaluate InSAR correction workflows in controlled scenarios with known ground truth; (2) to develop training sets and generative methods for machine learning algorithms; and (3) to educate on InSAR and its principles.
- Open Access
Engineering Reliable Deep Learning Systems
Recent progress in artificial intelligence (AI) using deep learning techniques has triggered its wide-scale use across a broad range of applications. These systems can already perform tasks such as natural language processing of voice and text, visual recognition, question-answering, recommendations and decision support. However, at the current level of maturity, the use of an AI component in mission-critical or safety-critical applications can have unexpected consequences. Consequently, serious concerns about reliability, repeatability, trust, and maintainability of AI applications remain. As AI becomes pervasive despite its shortcomings, more systematic ways of approaching AI software development and certification are needed. These fundamental aspects establish the need for a discipline on "AI Engineering". This paper presents the current perspective of relevant AI engineering concepts and some key challenges that need to be overcome to make significant progress in this important area.
Open AccessEngineering Reliable Deep Learning Systems
Recent progress in artificial intelligence (AI) using deep learning techniques has triggered its wide-scale use across a broad range of applications. These systems can already perform tasks such as natural language processing of voice and text, visual recognition, question-answering, recommendations and decision support. However, at the current level of maturity, the use of an AI component in mission-critical or safety-critical applications can have unexpected consequences. Consequently, serious concerns about reliability, repeatability, trust, and maintainability of AI applications remain. As AI becomes pervasive despite its shortcomings, more systematic ways of approaching AI software development and certification are needed. These fundamental aspects establish the need for a discipline on "AI Engineering". This paper presents the current perspective of relevant AI engineering concepts and some key challenges that need to be overcome to make significant progress in this important area.
- Open Access
- 26 December 2018
- Published by American Geophysical Union (AGU) in Journal of Geophysical Research: Space Physics
- Vol. 123 (12) , 10358-10367
Characterizing the Complex Two N‐Wave Ionospheric Signature of the 2016 Kaikoura Earthquake
Major earthquakes (>∼6.5 Mw) can generate observable waves which propagate not only through the Earth but also through the Earth's ionosphere. These traveling ionospheric disturbances can be observed using multifrequency GPS receivers to measure the ensuing perturbations in the Total Electron Content of the ionosphere. Assisted by a statistical approach we developed to indicate the occurrence of a significant TEC perturbation from the normal background behavior, we detect a traveling ionospheric disturbance generated by the 2016 7.8 Mw Kaikoura earthquake occurring in New Zealand on the 13th of November. The disturbance was detected ∼8 min after the earthquake, propagating toward the equator at ∼1 km/s with a peak‐to‐peak amplitude of ∼0.22 Total Electron Content units. The coseismic waveform exhibits complex structure unlike that of the expected N‐wave for coseismic ionospheric disturbances, with observations of oscillations with 4‐min periodicity and of two N‐waves. This observed complexity in the ionosphere likely reflects the impact of the complex, multifault structure of the earthquake.
- 26 December 2018
- Published by American Geophysical Union (AGU) in Journal of Geophysical Research: Space Physics
- Vol. 123 (12) , 10358-10367
Open AccessCharacterizing the Complex Two N‐Wave Ionospheric Signature of the 2016 Kaikoura Earthquake
Major earthquakes (>∼6.5 Mw) can generate observable waves which propagate not only through the Earth but also through the Earth's ionosphere. These traveling ionospheric disturbances can be observed using multifrequency GPS receivers to measure the ensuing perturbations in the Total Electron Content of the ionosphere. Assisted by a statistical approach we developed to indicate the occurrence of a significant TEC perturbation from the normal background behavior, we detect a traveling ionospheric disturbance generated by the 2016 7.8 Mw Kaikoura earthquake occurring in New Zealand on the 13th of November. The disturbance was detected ∼8 min after the earthquake, propagating toward the equator at ∼1 km/s with a peak‐to‐peak amplitude of ∼0.22 Total Electron Content units. The coseismic waveform exhibits complex structure unlike that of the expected N‐wave for coseismic ionospheric disturbances, with observations of oscillations with 4‐min periodicity and of two N‐waves. This observed complexity in the ionosphere likely reflects the impact of the complex, multifault structure of the earthquake.
- Open Access
- 9 December 2018
- Published by Wiley in ESS Open Archive
Transfer Learning of InSAR Atmospheric Effects for Earthquake Characterization
Interferometric synthetic aperture radar (InSAR) measures surface deformation from repeated passes of satellites or aircraft and has become an important tool to study geophysical phenomena such as earthquakes. However, InSAR data analysis is challenging due to atmospheric water vapor that can mimic the effects of Earth deformation and thus lead to wrong interpretations. We present preliminary results on how to differentiate between tropospheric effects and surface deformation from earthquakes using a convolutional neural network. As earthquake training sets are sparse, our approach leverages transfer learning techniques for tropospheric patterns from areas where deformations are known to be mostly absent over short time periods, and classifies specific areas of interferograms to reflect regions that are dominated by deformation or tropospheric noise. The applicability of the training set to a new area may depend on the similarity of the two climates. Examples of tropospheric delays are shown from interferograms constructed from Sentinel-1 data over part of southern California with short temporal baselines, and examples of deformation are taken from interferograms generated using Okada models. Our classifier is tested on data from the 2018 Oaxaca earthquake in Mexico from Sentinel-1. This work is a step towards using neural networks for a fine-granular tile-based validation of interferograms and automatically removing unwanted effects from InSAR signals, as well as towards enhancing the agility of disaster response programs. The open source code is available in the PyInSAR package on GitHub under the MIT license. We acknowledge support from NASA AIST80NSSC17K0125 (PI Pankratius) and NSF ACI1442997 (PI Pankratius).
- 9 December 2018
- Published by Wiley in ESS Open Archive
Open AccessTransfer Learning of InSAR Atmospheric Effects for Earthquake Characterization
Interferometric synthetic aperture radar (InSAR) measures surface deformation from repeated passes of satellites or aircraft and has become an important tool to study geophysical phenomena such as earthquakes. However, InSAR data analysis is challenging due to atmospheric water vapor that can mimic the effects of Earth deformation and thus lead to wrong interpretations. We present preliminary results on how to differentiate between tropospheric effects and surface deformation from earthquakes using a convolutional neural network. As earthquake training sets are sparse, our approach leverages transfer learning techniques for tropospheric patterns from areas where deformations are known to be mostly absent over short time periods, and classifies specific areas of interferograms to reflect regions that are dominated by deformation or tropospheric noise. The applicability of the training set to a new area may depend on the similarity of the two climates. Examples of tropospheric delays are shown from interferograms constructed from Sentinel-1 data over part of southern California with short temporal baselines, and examples of deformation are taken from interferograms generated using Okada models. Our classifier is tested on data from the 2018 Oaxaca earthquake in Mexico from Sentinel-1. This work is a step towards using neural networks for a fine-granular tile-based validation of interferograms and automatically removing unwanted effects from InSAR signals, as well as towards enhancing the agility of disaster response programs. The open source code is available in the PyInSAR package on GitHub under the MIT license. We acknowledge support from NASA AIST80NSSC17K0125 (PI Pankratius) and NSF ACI1442997 (PI Pankratius).
- 24 September 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 30 (4) , 766-777
A Hybrid Approach for Optimizing Parallel Clustering Throughput using the GPU
We introduce Hybrid-DBSCAN, that uses the GPU and CPUs for optimizing clustering throughput. The main idea is to exploit the memory bandwidth on the GPU for fast index searches, and optimize data transfers between host and GPU, to alleviate the potential negative performance impact of the PCIe interconnect. We propose and compare two GPU kernels that exploit grid-based indexing schemes to improve neighborhood search performance. We employ a batching scheme for host-GPU data transfers to obviate limited GPU memory, and exploit concurrent operations on the host and GPU. This scheme is robust with respect to both sparse and dense data distributions and avoids buffer overflows that would otherwise degrade performance. We evaluate our approaches on ionospheric total electron content datasets as well as intermediate-redshift galaxies from the Sloan Digital Sky Survey.
- 24 September 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 30 (4) , 766-777
A Hybrid Approach for Optimizing Parallel Clustering Throughput using the GPU
We introduce Hybrid-DBSCAN, that uses the GPU and CPUs for optimizing clustering throughput. The main idea is to exploit the memory bandwidth on the GPU for fast index searches, and optimize data transfers between host and GPU, to alleviate the potential negative performance impact of the PCIe interconnect. We propose and compare two GPU kernels that exploit grid-based indexing schemes to improve neighborhood search performance. We employ a batching scheme for host-GPU data transfers to obviate limited GPU memory, and exploit concurrent operations on the host and GPU. This scheme is robust with respect to both sparse and dense data distributions and avoids buffer overflows that would otherwise degrade performance. We evaluate our approaches on ionospheric total electron content datasets as well as intermediate-redshift galaxies from the Sloan Digital Sky Survey.
- Open Access
- 31 August 2018
- Published by American Geophysical Union (AGU) in Earth and Space Science
- Vol. 5 (8) , 393-407
Computer‐Aided Exploration of the Martian Geology
Motivated by growing amounts of data and enhanced resolution from orbiters and rovers, systems for computer‐aided decision support are becoming invaluable in planetary exploration. This article illustrates the value of such systems for a case study on the exploration of the Martian geology, along with improvements in assessing the favorability for landing. Under the current technical status quo for landing and rover's mobility, results show that Eastern Margaritifer Terra and Meridiani Planum stand out due to their high density of scientific targets and flat surfaces. However, our approach allows us to scale the analysis using different scenarios for the entire planet, quantifying the substantial benefits should higher landing elevations and higher rover speeds be realized in the future. This analysis offers new insights into the interplay of technical and scientific constraints.
- 31 August 2018
- Published by American Geophysical Union (AGU) in Earth and Space Science
- Vol. 5 (8) , 393-407
Open AccessComputer‐Aided Exploration of the Martian Geology
Motivated by growing amounts of data and enhanced resolution from orbiters and rovers, systems for computer‐aided decision support are becoming invaluable in planetary exploration. This article illustrates the value of such systems for a case study on the exploration of the Martian geology, along with improvements in assessing the favorability for landing. Under the current technical status quo for landing and rover's mobility, results show that Eastern Margaritifer Terra and Meridiani Planum stand out due to their high density of scientific targets and flat surfaces. However, our approach allows us to scale the analysis using different scenarios for the entire planet, quantifying the substantial benefits should higher landing elevations and higher rover speeds be realized in the future. This analysis offers new insights into the interplay of technical and scientific constraints.