AI/ML Research Scientist
I am currently working in the AI Lab at British Antarctic Survey developing a Polar Route planner applied to the RRS Sir David Attenborough (SDA) research vessel. I have a background in Geophysics applying a wide range of machine learning techniques for computational applications to real world problems.
In 2021 I completed an invited 2-year postdoctoral researcher position at California Institute of Technology investigating dynamics of earthquakes using cutting edge machine learning techniques. This work entailed computationally efficient research able to detect, associate and locate earthquake signals across 100’s of TerraBytes of continuous realtime data. The work completed throughout this position combined a wide variety of machine learning and statistics based techniques, including but not limited to: Physics Informed neural networks, Neural Operators, Stein Variational Inference, Normalizing Flows, Recurrent Neural Networks, Convolutional Neural Networks, and conventional Deep Neural Networks. In addition, during this position I helped supervise students from the Computer Science and Earth Science for undergraduate, postgraduate and PhD projects.
I received my Doctorate of Philosophy in Geophysics from Hughes Hall, University of Cambridge, investigating ‘Geomechanical properties of the Groningen gas reservoir’. This work combined extensive satellite and surface based remote sensing datasets (InSAR, cGPS and Optical Levelling), with regional earthquake catalogues, to better understand the surface and subsurface dynamics of active reservoirs; with the scope of forecasting the hazard of future production scenarios. During this position I also worked on volcanic hazard of central Iceland, providing infield knowledge of seismic network deployment, helping supervise masters student projects, and developing exhibits with demonstration for the 2016 Explosive Earth Royal Society Summer Student Exhibition.
I completed my undergrad in 2015 with a 2:i Masters of Earth Sciences from St Edmund Hall, University of Oxford.
Jonathan D. Smith, Samuel Hall, George Coombs, James Byrne, Michael A. S. Thorne, J. Alexander Brearley, Derek Long, Michael Meredith, Maria Fox, (2022, in review), Autonomous Passage Planning for a Polar Vessel, arXiv, https://arxiv.org/abs/2209.02389
Hadrien Meyer, Jonathan D. Smith and Jean-Philippe Avouac (in press), An integrated framework for surface deformation modeling and induced seismicity forecasting due to reservoir operations
Jonathan D. Smith, Elías R. Heimisson, Stephen Bourne and Jean-Philippe Avouac (2022), Stress-based forecasting of induced seismicity with instantaneous earthquake failure functions: Applications to the Groningen Gas Reservoir, Earth and planetary science letters https://doi.org/10.1016/j.epsl.2022.117697
Elías R. Heimisson, Jonathan D. Smith, Jean-Philippe Avouac and Stephen Bourne (2022), Coulomb Threshold Rate-and-State Model for Fault Reactivation: Application to induced seismicity at Groningen, Geophysical Journal International, https://doi.org/10.1093/gji/ggab467
Bing Q. Li, Jonathan D. Smith, and Zachary E. Ross (2021), Basal nucleation of ascending swarms in Long Valley Caldera, Science Advances doi
Jonathan D. Smith, Zachary E. Ross, Kamyar Azizzadenesheli and Jack Muir (2021), HypoSVI: Hypocentral inversion with stein variational inference and physics informed neural networks, Geophysical Journal International, doi:10.1093/gji/ggab309, https://arxiv.org/abs/2101.03271
Jean-Philippe Avouac, Maxime Vrain, Taeho Kim, Jonathan D. Smith, Thomas Ader, Elías R. Heimisson, Zachary Ross, Tero Saarno (2020), A convolution model for earthquake forecasting derived from seismicity recorded during the ST1 geothermal project on Otaniemi campus, Finland, Proceedings World Geothermal Congress 2020
Jonathan D. Smith, Kamyar Azizzadenesheli and Zachary E. Ross (2020), EikoNet: Solving the Eikonal equation with Deep Neural Networks, IEEE Transactions on Geoscience and Remote Sensing, arxiv:https://arxiv.org/abs/2004.00361
Zachary E. Ross, Elizabeth S. Cochran, Daniel T. Trugman and Jonathan D. Smith (2020), 3D fault architecture controls the dynamism of earthquake swarms, Science, doi:0.1126/science.abb0779
Jonathan D. Smith, Robert S. White, Jean-Philippe Avouac, and Stephen Bourne (2020), Probabilistic earthquake locations for induced seismicity in the Groningen region, Geophysical Journal International, doi:10.1093/gji/ggaa179
Jonathan D. Smith, Jean-Philippe Avouac, Robert S. White, Alex Copley, Adriano Gualandi and Stephen Bourne (2019), Reconciling reservoir compaction and compressibility in the Groningen region, Journal of Geophysical Research, doi: 10.1029/2018JB016801
Rebecca K. Pearce, Almudena Sánchez de la Muela, Max Moorkamp, James Hammond, Thomas M Mitchell, José Cembrano, Jaime Araya Vargas, Philip G Meredith, Pablo Iturrieta, Nicólas Pérez-Estay, Neill Marshall, Gonzalo Yañez, Ashley Griffith, Carlos Marquardt, Jonathan D. Smith, Ashley Stanton-Yonge, Rocio Núñez (2019), Interaction between hydrothermal fluids and fault systems in the in the Southern Andes revealed by magnetotelluric and seismic data, Earth and Space Science Open Archive, doi: 10.1002/essoar.10501143.1
Keije Chen, Jonathan D. Smith, Jean-Philippe Avouac, Zhen Liu, Y. Tony Song and Adriano Gualandi (2019), Triggering of the Mw 7.2 Hawaii earthquake of May 4, 2018 by a dike intrusion, Geophysical Research Letters, doi:10.1029/2018GL081428
Thomas S. Hudson, Jonathan D. Smith, Alex M. Brisbourne and Robert S. White (2019), Automated detection of basal icequakes and discrimination from surface crevassing, Annals of Glaciology, doi:10.1017/aog.2019.18
Ongoing project – Started Jan 2022
AI and Digital twinning to achieve carbon reduction on-board RRS Sir David Attenborough (SDA)
EikoNet: Solving the Eikonal equation with Physics Informed Neural Networks
Work Completed – [2004.00361] EikoNet: Solving the Eikonal equation with Deep Neural Networks (arxiv.org). Ulvetanna/EikoNet: Machine learning formulation for the Factored Eikonal Equation (github.com) . EikoNet: Solving the Eikonal Equation With Deep Neural Networks | IEEE Journals & Magazine | IEEE Xplore
The recent deep learning revolution has created enormous opportunities for accelerating compute capabilities in the context of physics-based simulations. Here, we propose EikoNet, a deep learning approach to solving the Eikonal equation, which characterizes the first-arrival-time field in heterogeneous 3D velocity structures. Our grid-free approach allows for rapid determination of the travel time between any two points within a continuous 3D domain. These travel time solutions are allowed to violate the differential equation—which casts the problem as one of optimization—with the goal of finding network parameters that minimize the degree to which the equation is violated. In doing so, the method exploits the differentiability of neural networks to calculate the spatial gradients analytically, meaning the network can be trained on its own without ever needing solutions from a finite difference algorithm. EikoNet is rigorously tested on several velocity models and sampling methods to demonstrate robustness and versatility. Training and inference are highly parallelized, making the approach well-suited for GPUs. EikoNet has low memory overhead, and further avoids the need for travel-time lookup tables. The developed approach has important applications to earthquake hypocenter inversion, ray multi-pathing, and tomographic modeling, as well as to other fields beyond seismology where ray tracing is essential.
HypoSVI: Hypocenter inversion with Stein Variational Inference and Physics Informed Neural Networks
Work Completed – HypoSVI: Hypocentre inversion with Stein variational inference and physics informed neural networks | Geophysical Journal International | Oxford Academic (oup.com) . [2101.03271] HypoSVI: Hypocenter inversion with Stein variational inference and Physics Informed Neural Networks (arxiv.org). Ulvetanna/HypoSVI: Hypocentral earthquake location using Stein-variational gradient descent and deep neural network travel-time formulations (github.com)
In this publication we introduce a scheme for probabilistic hypocenter inversion with Stein variational inference. Our approach uses a differentiable forward model in the form of a physics informed neural network (PINN), which we train to solve the Eikonal equation (see EikoNet project for more info). This allows for rapid approximation of the posterior by iteratively optimizing a collection of particles against a kernelized Stein discrepancy. We show that the method is well-equipped to handle highly multimodal posterior distributions, which are common in hypocentral inverse problems. A suite of experiments is performed to examine the influence of the various hyperparameters. Once trained, the method is valid for any seismic network geometry within the study area without the need to build travel time tables. We show that the computational demands scale efficiently with the number of differential times, making it ideal for large-N sensing technologies like Distributed Acoustic Sensing. The techniques outlined in this manuscript have considerable implications beyond just ray-tracing procedures, with the work flow applicable to other fields with computationally expensive inversion procedures such as full waveform inversion.
Seeing double: Digital twins and net zero
Blog 5 July, 2022