I successfully defended my thesis titled, 'Uncertainty Quantification, Knowledge Transfer, and Model Interpretability in Astronomy: A Machine Learning-Based Perspective' under the supervision of Prof. Desika Narayanan in the Department of Astronomy at the University of Florida, USA. I am interested in several areas of machine learning and deep learning, and find both theoretical developments and practical applications fascinating. I am especially excited about the areas of explainable AI, out-of-distribution identification, and uncertainty quantification; getting predictive uncertainties right and understanding the inner workings of ML models are no longer tasks that are just 'nice', but downright crucial if we want to get our models out of the lab and into the real world.
Astronomy makes available an enviable corpus of challenging, heterogenous, multi-modal datasets, filled with noisy, non-IID samples containing missing values. This presents a fantastic opportunity to validate and improve state-of-the-art modeling approaches in ML. I am working with a number of international collaborations to leverage these and create deployable, open-source stacks for solving some interesting problems: improving SED-fitting, automating the operation of telescopes, and predicting sky background-seeing. I am also working with collaborators to develop a new method of post-hoc model calibration (fixing the statistical properties of uncertainty predictions) and data augmentation.
I have decided to take a hiatus from astronomy and head to industry, where I look forward to engaging in bleeding-edge research in machine learning, and becoming a better coder.Profile Picture – adass-xxxi-2021/question_uploads/sankalp_profile_pic_qvegMCF.jpeg Affiliation –
University of FloridaPosition –
Graduate AssistantTwitter handle –
@spaceman_gildaGitHub ID –
We leverage state-of-the-art machine learning methods and a decade's worth of archival data from the Canada-France-Hawaii Telescope (CFHT) to predict observatory image quality (IQ) from environmental conditions and observatory operating parameters. Specifically, we develop accurate and interpretable models of the complex dependence between data features and observed IQ for CFHT's wide field camera, MegaCam. Our contributions are several-fold. First, we collect, collate and reprocess several disparate data sets gathered by CFHT scientists. Second, we predict probability distribution functions (PDFs) of IQ, and achieve a mean absolute error of ∼0.07′′ for the predicted medians. Third, we explore data-driven actuation of the 12 dome ``vents'', installed in 2013-14 to accelerate the flushing of hot air from the dome. We leverage epistemic and aleatoric uncertainties in conjunction with probabilistic generative modeling to identify candidate vent adjustments that are in-distribution (ID) and, for the optimal configuration for each ID sample, we predict the reduction in required observing time to achieve a fixed SNR. On average, the reduction is ∼15%. Finally, we rank sensor data features by Shapley values to identify the most predictive variables for each observation. Our long-term goal is to construct reliable and real-time models that can forecast optimal observatory operating parameters for optimization of IQ. Such forecasts can then be fed into scheduling protocols and predictive maintenance routines. We anticipate that such approaches will become standard in automating observatory operations and maintenance by the time CFHT's successor, the Maunakea Spectroscopic Explorer (MSE), is installed in the next decade.