Research : Deep Learnings and Cosmic Patterns

1. Installing Keras and basic examples

See this github wiki : https://github.com/shongscience/Keras-in-cosmology/wiki

2. Recognizing 3D patterns (shapes)

Results of The Toy Model

Ok.. the naive approach works great for the toy models, though there is an overfit threshold of the number of points. Simply, the increment of randomness by putting more points causes this overfit.

This can give some implications when we pool down the features, extracted from these point clouds.

Cosmic vs. Random

A simple MLP with naive coordinate values can give a 90% accuracy for classifying these two kinds of point clouds.

So... this is the limitation of the featureless MLP approach. To break this 90% barrier, we need to explore various feature extractions with new more complex NN baselines.

My guess is to combine CNN-ish Networks with Healpix-ish Convolution Kernels.

3. Any Practical Applications?

This point-classifier can recognize objects formed by points. As we have seen "drone shows" in Pyeongchang Olympic Games, this NN can recognize "drone formations". If this approach is applied to GANs (Generative Adversarial Networks), we will see drone shows, designed by A.I., not by humans.

3. Horovod : Distributed Deep Learning Platform

Running Horovod in Spark? Yes... the project hydrogen will make this possible.