Counting the Stars
summary
Project constructs a neural network that classifies Gaia variable stars using pandas numpy Keras (TensorFlow) and scikit‑learn on an IBM Power AC922 / Watson Machine Learning Accelerator, training on ~4.5k labeled stars and expecting ~300k next. Initial run reaches about 80% accuracy, after removing correlated features and using dropout 50% plus randomized search then local grid search it improves to 94% accuracy. PCA is tested then abandoned.
AI-written summary · about
chapters (15)
- 0:00Intro
- 3:46Variable Stars
- 5:37w150 Attributes per star
- 6:56What do the data look like?
- 9:25Cleaning data
- 11:32Two phases of machine learning
- 12:38Approach
- 13:35Frameworks
- 15:08Minimal network - clean the data
- 18:07Minimal network - build the neural network
- 19:38Overfitting
- 20:41Learning parameters
- 25:10Optimisation
- 28:56Future work
- 30:49Lessons learnt
related talks
39mData, Science: Investigating 1 Million Galaxies with Humans and TensorFlow
36mTwinkle twinkle little star, how I wonder what you are
45mData Mining in Astronomical Survey Data of Variable Stars with Python
44mSidharth Kumar of Flipkart at Machine Learning Developers Summit 2019
23mUsing a Bayesian Neural Network in the Detection of Exoplanets
38mIs There A Needle In The Haystack?
53mFROM ZERO TO DEEP LEARNING WITH SCALA
38mReal Life Customer Stories with AI