CMCL’s new preprint presents the results of the application of Deep Kernel Learning (DKL) to a modern Diesel fuelled Compression Ignition Engine. DKL is a state-of-the-art machine learning technique that combines a deep feedforward neural network with the flexibility of a non-parametric Gaussian Process regression.

The DKL feature within CMCL’s Model Development Suite (MoDS) software offers estimation of the underlying hyperparameters, and in this case is applied to formulate robust NOx and soot emissions surrogate models. The performance of the resulting DKL surrogate is also compared with a plain neural network technique, a plain Gaussian process as well as a High Dimensional Model Representation (HDMR).

You can see the preprint on our Publications page, or download it directly here.