Neural+Networks

In neural networks, our base element is a "neuron." These neurons each have input and output connections. Typically, our network is designed in three layers: input, intermediate (hidden layers), and output. This type of layout allows for the network to not only follow instructions, but also to find familiar patterns data sets!

According to Pinheiro et. al. in "Application of Neural Networks to the study of stellar model solutions," " ANNs have been used to perform an automated classification of stellar spectra and determine global stellar parameters (luminosity L, effective temperature Teff, and metal abundance [M/H]) from low resolution spectra" [2] Their major motivation for undertaking the use of ANNs was to add other parameters to the list including stellar age, helium abundance, metalicity... Finding a model fit for a system with 4+ variables is no easy task for a standard computer system, so they turned to Neural Networks!

The reasons they give for using neural networks over traditional technologies are four fold:


 * 1) Adaptive Learning: An ability to learn how to do tasks based on the data given for training or initial experience.
 * 2) Self-Organization: An NN can create its own organization or representation of the information it receives during learning time.
 * 3) Real Time Operation: NN computations may be carried out in parallel, and special hardware devices are being designed and manufactured which take advantage of this capability.
 * 4) Fault Tolerance via Redundant Information Coding: Partial destruction of a network leads to the corresponding degradation of performance. However, some network capabilities may be retained even with major network damage.

Having gone through their initial data set with success, essentially showing proof-of-concept, they plan to apply their ANN to a larger sample size, even expanding their research to an interactive database online! This partially captures the spirit of the Zooniverse community, allowing users to add to the ever expanding collection of knowledge.

Similar successes have been achieved in stellar classification by other groups around the world--from India [3] to Spain [4].

However, the studies are not limited to just stellar classification. As shown in Eatough et al. [5], radio pulsars can be identified using ANNs as well! If that's the case, then studies involving obtaining flux values (brightness) from stars can utilize ANNs as well. These potential candidates include exoplanet searches and the identification/studying of variable stars.