Simulations of biophysical systems have provided a huge contribution to our fundamental understanding of human physiology and remain a central pillar for developments in medical devices and human machine interfaces. However, despite their successes, such simulations usually rely on highly computationally expensive numerical modelling, which is often inefficient to adapt to new simulation parameters. This limits their use in simulating dynamic human behaviours, which typically proceed along a sequence of small time steps. One may painstakingly produce a few static simulations at discretised stages, but not the hundreds of simulations that are essential to capture the dynamic nature of human body. We propose that an alternative approach is to use conditional generative models, which can learn complex relationships between the underlying generative conditions and the output data whilst remaining inexpensive to sample from. As a demonstration of this concept, we present BioMime, a hybrid-structured generative model that combines elements of deep latent variable models and conditional adversarial training. We demonstrate that BioMime can learn to accurately mimic a complex numerical model of human muscle biophysics and then use this knowledge to continuously sample from a dynamically changing system in a short time. This ultimately converts a static model into a dynamic one with no effort. We argue that transfer learning approaches with conditional generative models are a viable solution for dynamic simulation with any numerical model.

翻译：生物物理系统的模拟极大地促进了我们对人体生理学的基础理解，并始终是医疗器械和人机接口发展的核心支柱。然而，尽管取得了成功，这类模拟通常依赖于计算成本极高的数值建模，且难以高效适应新的模拟参数。这限制了其在模拟人体动态行为中的应用——此类行为通常沿着一系列微小时间步长展开。人们可能费尽心力在离散阶段生成少量静态模拟，却无法获得捕捉人体动态本质所必需的数百次模拟。我们提出一种替代方案：采用条件生成模型，这类模型能够学习底层生成条件与输出数据之间的复杂关系，同时保持采样的低成本。为论证这一概念，我们提出BioMime——一种结合深度潜变量模型与条件对抗训练要素的混合结构生成模型。我们证明，BioMime能够学习精确模仿人体肌肉生物物理的复杂数值模型，并利用该知识在短时间内持续对动态变化系统进行采样。这最终可将静态模型无需额外努力地转化为动态模型。我们认为，基于迁移学习的条件生成方法适用于任何数值模型的动态模拟。