This will be done by implementing an original idea for a multichannel neural network, which would have the ability to take both the Sunspotter SMART detection values along with the corresponding images. The Autoencoder network from #19 will be used to get an effective lower-dimensional encoding for the images. This shall be concatenated and re-normalised with the processed SMART detection values and the complexity score to make the final feed-forward neural network that will learn the mapping.
We shall also retrain the best performing non-Deep Learning model to see if we can get comparable results from less computationally taxing algorithms.
This will be done by implementing an original idea for a multichannel neural network, which would have the ability to take both the Sunspotter SMART detection values along with the corresponding images. The Autoencoder network from #19 will be used to get an effective lower-dimensional encoding for the images. This shall be concatenated and re-normalised with the processed SMART detection values and the complexity score to make the final feed-forward neural network that will learn the mapping.
We shall also retrain the best performing non-Deep Learning model to see if we can get comparable results from less computationally taxing algorithms.