Browsing by Author "Simmons, M. J. H."
Now showing 1 - 2 of 2
Results Per Page
Sort Options
Item Open Access The duality between particle methods and artificial neural networks(Nature Publishing Group / Nature Research, 2020-10-01) Alexiadis, Alessio; Simmons, M. J. H.; Stamatopoulos, K.; Batchelor, H. K.; Moulitsas, IreneThe algorithm behind particle methods is extremely versatile and used in a variety of applications that range from molecular dynamics to astrophysics. For continuum mechanics applications, the concept of ‘particle’ can be generalized to include discrete portions of solid and liquid matter. This study shows that it is possible to further extend the concept of ‘particle’ to include artificial neurons used in Artificial Intelligence. This produces a new class of computational methods based on ‘particle-neuron duals’ that combines the ability of computational particles to model physical systems and the ability of artificial neurons to learn from data. The method is validated with a multiphysics model of the intestine that autonomously learns how to coordinate its contractions to propel the luminal content forward (peristalsis). Training is achieved with Deep Reinforcement Learning. The particle-neuron duality has the advantage of extending particle methods to systems where the underlying physics is only partially known, but we have observations that allow us to empirically describe the missing features in terms of reward function. During the simulation, the model evolves autonomously adapting its response to the available observations, while remaining consistent with the known physics of the system.Item Open Access The virtual physiological human gets nerves! How to account for the action of the nervous system in multiphysics simulations of human organs(The Royal Society, 2021-04-14) Alexiadis, Alessio; Simmons, M. J. H.; Stamatopoulos, K.; Batchelor, H. K.; Moulitsas, IreneThis article shows how to couple multiphysics and artificial neural networks to design computer models of human organs that autonomously adapt their behaviour to environmental stimuli. The model simulates motility in the intestine and adjusts its contraction patterns to the physical properties of the luminal content. Multiphysics reproduces the solid mechanics of the intestinal membrane and the fluid mechanics of the luminal content; the artificial neural network replicates the activity of the enteric nervous system. Previous studies recommended training the network with reinforcement learning. Here, we show that reinforcement learning alone is not enough; the input– output structure of the network should also mimic the basic circuit of the enteric nervous system. Simulations are validated against in vivo measurements of high-amplitude propagating contractions in the human intestine. When the network has the same input–output structure of the nervous system, the model performs well even when faced with conditions outside its training range. The model is trained to optimize transport, but it also keeps stress in the membrane low, which is exactly what occurs in the real intestine. Moreover, the model responds to atypical variations of its functioning with ‘symptoms’ that reflect those arising in diseases. If the healthy intestine model is made artificially ill by adding digital inflammation, motility patterns are disrupted in a way consistent with inflammatory pathologies such as inflammatory bowel disease.