A European Space Agency research group uses Spectral Energy Distribution (SED) to analyze their proto-solar systems. They have a very well-defined and complex mathematical system in place that can provide accurate results regarding the existence of the crucial elements needed for existing life. Each analysis takes 52 hours to compute, and there are hundreds of parameters that must be tested.
The client was looking for a way to improve the performance of this task and enhance their opportunities for investigation.
To enhance this process, we decided to implement a Deep Learning approach. Deep Learning technologies have the ability to approximate expected results given a specific input that can accurately mimic their hyper-complex mathematical system.
Deep Learning technologies have the ability to mimic complex systems, including those that are heavily based on physics and chemistry, meaning that they can be applied to a huge variety of applications. Because of its diverse range of applications, Deep Learning can have a significant impact on processes when used within industries such as pharmaceuticals, aeronautics, energy providers, and many more. It’s for this reason decided that Deep Learning would be best applied to this case.
We trained several Deep Learning models using the data from the inputs and results of past experiments. Once the Deep Learning models had learned enough, we tested the new SED datasets and checked for both performance and accuracy. After several rounds of testing and fine tuning, we chose the most balanced Deep Learning model for the Astrophysicists to start using.
Their system took an average of 52 hours to run an experiment, the DL solution provided by Intelygenz was capable of running the same experiment in 5 hours with the same accurate result, reducing computing time by 90.4%.
After testing the Deep Learning solution in real scenarios Intelygenz pointed out to the Astrophysicists that the same Deep Learning model could be used in both ways in order to not only simulate their algorithms but also to propose experiment parameters in order to match with a desired SED.
This final solution helps Astrophysicists speed up their research thanks to the application of Deep Learning to optimize their complex systems.