Application of Machine learning to understand gas-liquid interactions in nanopores

The use of Machine Learning has spread to many fields, from commodity price prediction to agricultural management. The interest in machine learning often stems from the ability to iteratively optimise a computer's ability to perform a task without the need for explicitly programming the process. As a result, some degree of automation is achievable and human involvement may be minimised, benefiting productivity where expensive or repetitive tasks are involved.
Support Vector Machines (SVMs) are Machine Learning models which make use of algorithms to perform classification and regression analysis. To perform these operations the SVM may make use of a range of strategies, amongst which are Gaussian similarity kernels, which in principle are Gaussian Processes. A Gaussian Process is a stochastic process where between each data point there is a multivariate normal distribution, so the calculated data is taken at face value and the points between are predicted probabilistically along with their marginal distribution.
The SVM in this instance is applied to molecular potential energy surfaces. A potential energy surface (PES) describes the energy of a molecular system with respect to intermolecular distances.
Ab-initio computation of the intermolecular PES accurately emulates the electron clouds from established quantum principles. The downside is frequently the cost of lengthy calculations, which naturally leads efforts to interpolate smaller data sets to save computing time while still obtaining an accurate PES. The problem arising is that parametric interpolation is a flawed system, making use of a best fit which may not coincide with the true potential at a given point. As a result, machine learning methods such as SVM regression which respect the calculated data are gaining traction for PES interpolation.