skip to main content
Home  /  Undergraduate Research  /  Programs  /  Amgen Scholars  /  Announcements of Opportunity

Amgen Scholars: Announcements of Opportunity

Below are Announcements of Opportunity posted by Caltech faculty for the Amgen Scholars program.

Announcements of Opportunity are posted as they are received. Please check back regularly for new AO submissions! Remember: This is just one way that you can go about identifying a suitable project and/or mentor. For additional tips on identifying a mentor click here.

Please remember:

  • Students pursuing Amgen must be U.S. citizens, U.S. permanent residents, or students with DACA status.
  • Students pursuing Amgen must complete the 10-week program from June 18 - August 23, 2024. Students must commit to these dates. No exceptions will be made.
  • Accepted students must live in provided Caltech housing.

Showing Record 1 of 63    Next >>           Back To List

Project:  RAPID ML
Disciplines:  Data Science, Astronomy
Mentor:  Ashish Mahabal, Deputy Director, Center for Data Driven Discovery, (PMA),, Phone: 6263954201
Mentor URL:  (opens in new window)
Background:  The ROMAN telescope will be launched in May 2027. It will use a 2.4m telescope and observe a per image area of 0.28 sq. degrees. At Caltech the RAPID (Roman Alerts Promptly from Image Differencing) team will provide rapid image differencing, prompt alerts from the difference images, source-matched light curves for all Roman candidates, and forced photometry for Roman photometric history. For this to happen properly we will use machine learning. Until real data are available we will create tools using simulated data, and transients of different types injected into the images.
Description:  Transient events will be extracted from the point-source matched-filter signal-to-noise ratio images, where detection will be made on both the positive (science minus reference) and negative (reference minus science) images. Both aperture and PSF-fitted photometry, at the location of the events, will be performed. At this stage, association with the nearest source in the reference image will also be carried out with an ML score of whether this is a star or galaxy.
The first classification task is to develop a real-bogus ML score that robustly flags artifacts. Artifacts come in many forms: (a) cosmic ray hits, (b) saturated stars (c) reflections, (d) cross-talk, (e) other features originating in the electronics. Each instrument/telescope combination has different characteristics, and hence different set of artifacts that need to be characterized, and separated from genuine sources. Given the variety of exposure times for Roman surveys, and the IR wavelengths, there may be additional noise patterns. Level-2 Roman data products may already be rid of many artifacts (e.g., streaks from saturated stars and cosmic ray hits caught in the upstream pipeline). The plan is to capture any residual faint artifacts that may escape the first level by going deeper and using ML algorithms on detected sources, including a star-galaxy classification.
The next classification task is to further classify the real sources into the astrophysical source class. The real sources can be characterized and classified using additional ML models, both classical ML (e.g., random forest and XGBoost) and Deep Learning (e.g., Convolutional Neural Networks [CNNs], Long Short-Term Memory [LSTM] networks). Also consider more recent techniques like transformers, assess applicability, relevance and interpretability of the classifications. Each type would likely have some sub-types that need to be separated. This translates to a multi- class classifier, or multiple binary-classifiers. Either way, one needs a representative set of images for each type/subtype. Before launch, injected sources will be carefully selected to display a wide range of characteristics to build a robust model. After launch, during commissioning, retraining will be done to improve the model. As the available dataset grows, periodic retraining of the model will be done to optimize performance.

The ML results will be included in the alert packets as metadata for use downstream by as- tronomers interested in different facets of time-domain science.
ZTF tool development will be used as a guide for some of the architectures.
References:  Roman telescope:
ACAI filters:
Student Requirements:  Knowledge of python, jupyter notebooks (Google Colab), git, and databases like Mongo DB. Conversant with basics of machine learning and statistics, knowledge about linux/unix. Basic astronomy knowledge will be a plus.
Programs:  This AO can be done under the following programs:

  Program    Available To
       SURF    Caltech students only 

Click on a program name for program info and application requirements.

Showing Record 1 of 63    Next >>           Back To List