@inproceedings {pub4277,
	title = {Comparison of deep learning and hand crafted
features for mining simulation data},
	author = {Theodoros Georgiou AND Sebastian Schmitt AND Markus Olhofer AND Thomas B{\"a}ck AND Michael Lew},
	year = {2020},
	abstract = {Computational Fluid Dynamics (CFD) simulations are very important for a plethora of industrial applications, such as optimization of aerodynamic properties of different engineering designs, e.g. cars, airplanes etc.. The output of these simulations can become very complex and hard to interpret due to their high dimensionality (3D space + time dependence with more than six values per point in the space-time). There have been many works that try to automatically extract meaningful information from these kind of simulations. In this paper we (i) propose an adaptation of the classical hand crafted features from computer vision for the purpose of describing such content, (ii) adapt deep learning methods for this kind of data, (iii) construct a dataset of 2D simulations for the purpose of benchmarking methods and (iv) make a comparison of these methods in being able to accurately describe the content of CFD simulation output on the proposed dataset.
},
	publisher = {IEEE},
	booktitle = {International Conference on Pattern Recognition}
}