View Position
| Project Title | Machine learning applied to Cosmological Simulations |
| Summary | Cosmological simulations provide unique insights into the evolutionary history of our Universe and are becoming increasingly more important in quantifying the accuracy of cosmological measurements. We propose to accelerate this process by exploring the role machine learning can play in quickly generating mock galaxy catalogs from cosmological simulations that can be obtained orders of magnitude faster than by using traditional techniques. In addition, we will explore the combination of n-body+hydro simulations as training data for these machine learning algorithms to make more accurate mock galaxy catalogs than permitted by traditional techniques. |
| Job Description | The selected intern will apply a variety of supervised, semi-supervised, and unsupervised learning algorithms to existing cosmological simulation dark matter halo catalogs in order to rapidly create new mock galaxy catalogs. Traditionally semi-analytic methods are employed that can take significant time and computational resources to generate mock galaxy catalogs from existing dark matter halo catalogs. The selected student will use existing mock catalogs to train machine learning algorithms to make new mock catalogs that retain the same level of accuracy, yet can be generated much faster. Specific algorithms that we expect to explore include SVM, random forests, self-organizing Maps, and deep learning methods like convolution neural networks. These will initially be demonstrated by using the sickout_learn library, and as the student gains experience we will move to faster libraries or develop new techniques in C++. An additional possibility includes the extension of this work to leverage C++ meta-programming that can leverage GPUs in the XK-7 nodes. |
| Conditions/Qualifications | Student should be familiar with cosmological n-body simulations and basic machine learning tasks. |
| Start Date | 05/16/2015 |
| End Date | 05/31/2016 |
| Location | University of Illinois, Urbana, IL |
| Interns | Harshil Kamdar
|