3
$\begingroup$

I am working on a problem in which I am attempting to find a stable region in a spiral galaxy. The PI I'm working with asked me to use machine learning as a tool to solve the problem. I have created some visualizations of my data, as bellow.

data

In this image, you can see there is a flat region between 0 and roughly 30 pixels, and between 90 pixels and 110 pixels. I have received suggestions to use an RNN LSTM model that can identify flat regions, but I wanted to hear other suggestions of other neural network models as well.

The PI I'm working with suggests to feed my data visualization images into a neural network and have the neural network identify said stable regions. Can this be done using a neural network, and what resources would I have to look at? Moreover, can this problem be solved with RNN LSTM? I think the premise of this was to treat the radius as some temporal dimension. I've been extensively looking for answers online, and I cannot quite seem to find any similar examples.

$\endgroup$
0
$\begingroup$

In image processing CNNs are usually used to create weighted filters for focusing in on the image features which are most important for making predictions. Keras is one of the libraries used to examine images in this way. With this type of analysis you will need labeled and unlabeled data you want to create a network that inputs a photo extracts the flat black line regions and outputs those. The model will be generative, generating guesses of regions where the function is flat. This is all possible to do but in order to label the data you need to label them by hand or you need to create a function that manually labels them which would not be very difficult. The input nodes will take in the pixels of the picture and the output layer will be guesses at location along the graph of wether the section is flat or not. It seems overkill to do this with a neural network when it is possible to not use a NN and creating a labeling method will most likely be your first step. If you have any questions please ask.

$\endgroup$
0
$\begingroup$

If you're really just trying to find long contiguous flat regions in a sequence, you do not need machine learning. Your PI is mistaken. You would be better off simply writing a short data processing program. Your program could find the finite differences between adjacent datapoints, and then count whether a long string of them are below some threshold to identify long flat regions. This will be faster, simpler, and perhaps more accurate than using ML on data visualizations for this task.

If you are trying to find something more complex than these long flat regions, you could instead train an LSTM on the raw sequential data that you are using to generate the images. Again, that will probably be more accurate than trying to train a CNN, or any non-sequential model, on the image data itself.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.