Plenty of effort has been dedicated to move towards big data, machine learning and artificial intelligence (AI) techniques. Indeed, several industries have selected such techniques as a path to follow in the near/far future for their own research. The reader may think about recent Google’s artificial intelligence software release (Tensor Flow).
Another example quite impressive can be found in this short article,where Boeing industries claims that the Boeing 787 produces over 500 Gb of data during every flight.
Certainly, the standards behind AI are that the more data is available; the better training the system will receive, thus, increasing the capability of prediction. However, even if no one doubts of the performance of machine learning techniques, some collateral problems may appear as we analyze the situation. Specialists in the field cannot avoid questions such as: How is this data going to be stored? How is it going to be treated? Indeed, the more data is available the harder it becomes to handle it. This graph estimates the amount of data stored in the world, as it can be seen, the growth is exponential due to new technologies appearing in society such as Smartphones, high performance computers, etc.
It is clear that at some point the cloud will have problems to handle all this data. That is the reason why, the mankind has to do an exercise of introspection demanding if it is necessary to store such quantities of data.Certainly not, that is the reason why there are plenty of compression algorithms in order to alleviate such problematic. For instance, when you take a picture with your mobile phone, the mobile does not save every pixel that your camera is able to capture. Instead of doing that, there are some mathematical algorithm that does a change of coordinates making possible to store less coefficients in this new coordinate system, that is the reason why the extensions suchas Picture.JPEG or Picture.PNG among other exists. In this figure, it is shown a image which has been compressed up to 96% retaining the main features of the image.
Model Order Reduction algorithms share the same objective than those compression algorithms, since they are seeking the best coordinate system (the “best basis”) in which few coefficients are required in order to express a given object. There are techniques which guess this “best basis” using a precomputed set of data (or snapshoots) creating a family of a posteriori methods as POD or RB. Moreover, we have techniques which are able to find this “best basis” a priori like in the PGD case, creating the set of modes that best fit the problem. What it is clear is that once you are in the proper space, your model order reduction will be very efficient, just like your compression algorithm. In my personal opinion, such “best basis” contain the most important correlations hidden in the big data, transforming data into knowledge. Indeed, we do not have to store data, we have to store knowledge.
