Andreas Zoglauer, Staff Researcher

Closed (1) Pioneering new Data-Analysis Techniques for the next generation of gamma-ray space telescope with Machine Learning

Applications for spring 2021 are now closed for this project.

The next generation of gamma-ray space telescopes aims to open a new window into gamma-ray astronomy with unprecedented angular resolution and sensitivity. The goals of these new telescopes range from trying to detect dark matter, to better understand the element formation in our Galaxy, to identifying the physical processes at work in the extreme conditions around black holes.

However, to unlock their full potential, we need to develop new data analysis techniques. This project is centered on improving the existing (experimental) data-analysis pipeline with the latest machine learning techniques. The available topics include:
(1) Use various machine learning approaches to track the particles in our detectors such as graph convolutional neural networks.
(2) Near real-time localization of gamma-ray bursts.

All topics are novel and should - if we get it work - ultimately lead to published papers.

As a member of this URAP team, you can pick one machine learning topic and apply the TensorFlow / Keras machine learning library to it. The ultimate goal is to improve upon the currently existing reference implementations with new, cutting-edge machine learning approaches. You are expected to work in a team of 2-4 URAP students and meet with your team at least once during the week and work together on the project. In addition, regular Zoom meetings will be held with your supervisor. The data sets containing the test and training data will be provided.

The concrete learning goals depend on your chosen topic. In general, they include:
(1) Learn about the data analysis of a future gamma-ray telescopes
(2) Deepen your knowledge about TensorFlow / Keras
(3) Learn how to apply these tool kits to the data of a modern telescope

Since this is a multi-year project, we prefer students who want to work with us for more than 1 semester. The time spent on the project should be on average at least 6 hours per week, more strongly preferred.

Qualifications: Interest in astrophysics (in particular nucleosynthesis, dark matter, supernovae, black holes); full proficiency in Python; prior experience with machine learning with neural networks (e.g. CNN, G(C)NN) and with a common machine learning library such as TensorFlow / Keras preferred; familiarity with the Linux/Unix environment, ssh, bash, and git; great organization and communication skills; punctuality and reliability; talent for multitasking and balancing this project with your normal classes. Exceptions can be made if you are highly motivated.

Weekly Hours: to be negotiated

Off-Campus Research Site: During the fall 2020 semester, we will only meet and communicate via Zoom, Slack, and email.

Related website: https://github.com/zoglauer/gamma-ai

Closed (2) Improving the data-analysis pipeline of the COSI telescope with machine learning

Applications for spring 2021 are now closed for this project.

COSI, the Compton Spectrometer and Imager, is a NASA-funded balloon-borne gamma-ray telescope, observing Galactic nucleosynthesis, Galactic positron annihilation as well as the most violent events in our Universe (supernovae, neutron star mergers) and the most extreme environments (pulsars, black holes). COSI had a very successful 46-day balloon flight in 2016 starting from Wanaka, NZ, once around Antarctica, and landing in the Atacama Desert in Peru. Currently the analysis of this data is ongoing as well as the preparations for the next COSI flight in spring 2021 (or later depending on Covid-19).

This project is centered on improving the data analysis pipeline of COSI by applying the latest machine learning tools to individual segments of the pipeline. The available machine learning topics are:
(1) Improve an existing approach for the localization of the gamma-ray interactions in our Germanium detector with a neural network.
(2) Identify background events such as Earth Albedo events, incompletely absorbed events, wrongly reconstructed events and internal radioactive decays using machine learning on simulated training data sets.
(3) Approximate the imaging response using a deep neural network.

All topics are novel and should - if we get it to work - ultimately lead to published papers.


As a member of this URAP team, you can pick one machine learning topic and apply either the TensorFlow or the TMVA machine learning library to it. The ultimate goal is to improve upon the currently existing reference implementations with new, cutting-edge machine learning approaches. You can either work alone or in a team on your topic. The data sets containing the test and training data will be provided.

The concrete learning goals depend on your chosen topic. In general, they include:
(1) Learn about the data analysis of a modern NASA-sponsored telescope, COSI
(2) Learn how to use a machine learning toolkit: TensorFlow or TMVA
(3) Learn how to apply these tool kits to the data of a modern telescope

Since this is a multi-year project, we prefer students who want to work with us for more than 1 semester. The time spent on the project should be on average at least 6 hours per week, more strongly preferred. Regular meetings will be held via Zoom.


Qualifications: Data 8 or equivalent; proficiency in Python or C++; prior experience with machine learning (e.g. (deep) neural networks, (random forests of) (boosted) decision trees, support vector machines) and with a common machine learning library recommended (e.g. scikit-learn, tensorflow, keras, TMVA); familiarity with the Linux/Unix environment, ssh, bash, and git; interest in Physics and Astrophysics; great organization and communication skills; punctuality and reliability; talent for multitasking and balancing this project with your normal classes. Exceptions can be made if you are highly motivated.

Weekly Hours: to be negotiated

Off-Campus Research Site: During the fall 2020 semester, we will only meet and communicate via Zoom, Slack, and email.

Related website: http://cosi.ssl.berkeley.edu
Related website: https://github.com/zoglauer/gamma-ai