BIDS hosted the third annual Astro Hack Week during the week of August 29–September 2. Astro Hack Week is a summer school and collaborative workshop, the goal of which is to apply modern techniques in statistics, machine learning, and computing to problems in astronomy.
Astro Hack Week has a distinctive format: Monday through Thursday mornings consisted of in-depth lectures on topics in statistics and machine learning. While these mornings were typical of a summer school, the rest of the week (afternoons, evenings, and all day Friday) were broadly unstructured: the approximately 50 participants formed small groups and worked together on collaborative research projects, applying concepts covered during the morning sessions to real astronomy research problems.
Scattered throughout the afternoons were breakout sessions: optional ad hoc lectures on popular topics selected by popular vote by participants during the week. This year, we had breakouts on Gaussian Processes, databases, automatic differentiation, packaging and documentation in Python, Git and GitHub, and hierarchical Bayesian inference.
On Friday, one of the week’s organizers, Daniela Huppenkothen, gave the BIDS Data Science Lecture. In her talk, titled “Why Your Field Needs a Hack Week: Bringing Data Science Into Astronomy,” Daniela discussed the ideas behind Astro Hack Week and how similar concepts might be applied to other fields.
Finally, on Friday afternoon, we wrapped up the week by presenting the results of everyone’s projects during the week. There were far too many interesting projects to list here, but the following is a small sample:
- Using neural networks for classification of radio antenna fitness (https://github.com/zakiali/
radio_nnet) - Using gaussian processes to determine galaxy redshifts (https://github.com/aimalz/
densityfield) - Infering the 3-d shape of molecules or galaxies from many extremely low-count images (https://github.com/
davidwhogg/ DiffractionMicroscopy) - A tool for collaborators or teachers and learners to find each other (https://github.com/benelson/
collaboratr) - A tool for children to draw their own constellations (https://hackpad.com/New-
Constellations-QH2i2IHKFBG)
A collaboration between the three institutes comprising the Moore-Sloan Data Science Environments, Astro Hack Week has rotated between the three locations, with the 2014 and 2015 editions being held at the UW eSciences Institute and the NYU Center for Data Science, respectively. The concept has recently inspired similar hack weeks in other fields. The first Neuro Hack Week took place at the University of Washington’s eSciences Institute the first week of September. In November, the first Geo Hack Week will be held (also at the eSciences Institute).
Many thanks are due to the Moore and Sloan Foundations for sponsoring the event as well as to GitHub for generously hosting the entire group at its headquarters in downtown San Francisco on Wednesday evening.
All four morning lectures during the week were recorded in their entirety and are available on the BIDS YouTube channel:
