Wednesday Nov 13th 3 pm, 131 Campbell

A new generation of high etendue telescopes is allowing us to explore large volumes of the Universe with fast cadences. This has led to the discovery of new populations of events or the new phases of evolution of known populations of events. In order to take advantage of these new discoveries several new tools are required. Among them are high performance image processing tools, fast machine learning aided discovery and classification algorithms, interoperable tools that allow an effective communication between the different astronomical infrastructure, new models which allows interpreting new regions of the parameter space, and new tools to extract the most physical knowledge from these observations. In this talk I will review some examples of high cadence surveys, their tools and scientific results, in particular concerning our experience with the High cadence Transient Survey (HiTS). I will also discuss the future ecosystem of time domain astronomy in the era of high cadence observations, where a new layer of astronomical alert brokers and target and observation managers will be required to connect survey and follow-up telescopes. In particular, I will discuss the challenges and opportunities found while developing the ALeRCE astronomical broker and its implications for the future of multi messenger astronomy.

Oct 8th, 3pm, 501B Campbell

Neural networks are a central technique in machine learning. Recent years have seen a wave of interest in applying neural networks to physical systems for which the governing dynamics are known and expressed through differential equations. Two fundamental challenges facing the development of neural networks in physics applications is their lack of interpretability and their physics-agnostic design. The focus of the present work is to embed physical constraints into the structure of the neural network to address the second fundamental challenge. By constraining tunable parameters (such as weights and biases) and adding special layers to the network, the desired constraints are guaranteed to be satisfied without the need for explicit regularization terms. This is demonstrated on supervised and unsupervised networks for two basic symmetries: even/odd symmetry of a function and energy conservation. In the supervised case, the network with embedded constraints is shown to perform well on regression problems while simultaneously obeying the desired constraints whereas a traditional network fits the data but violates the underlying constraints. Finally, a new unsupervised neural network is proposed that guarantees energy conservation through an embedded symplectic structure. The symplectic neural network is used to solve a system of energy-conserving differential equations and out-performs an unsupervised, non-symplectic neural network.

May 28th, 3pm, 131 Campbell

Optical astronomy has seen several applications of deep learning where some structure in the inputs is discernible e.g images of galaxies and streaking asteroids. Much of time-domain astronomy relies on light-curves where identifying structure is the basis of classification and yet non-trivial at best. As a result classification with sparse light curves continues to be a hard problem. We show how convolutional networks can be used for effective classification of light curves. Recurrent Networks have also been employed with light curves, but recently there has been discussion as to how they may not be that effective, and convolutional networks of a different type are seen as a better alternative. We will speculate about where sparse light curve classification may be heading with these tools as ZTF is nearing the halfway point, and LSST looms large.

May 7th, 3pm, 131 Campbell

In this talk I will describe how to measure state-of-the-art stellar ages by combing information from different evolutionary processes, including hydrogen burning and magnetic braking. I will also describe the process of measuring stellar rotation periods, including why this is hard, all the various pitfalls, and how we can improve it. Finally, I will discuss how we might measure the ages and other properties of stars from time-series brightness variations of stars (light curves) directly, without using summary statistics to translate from light curve to age.

Mar 22nd, 2 pm, 131 Campbell

Upcoming large observational time-domain surveys such as the Large Synoptic Survey Telescope (LSST) and the Transiting Exoplanet Survey Satellite (TESS) will produce millions of regularly- and irregularly-sampled astronomical light curves. The enhanced sensitivity and time-sampling strategies of these surveys will open new discovery windows in diverse fields of astronomy, from exomoons to precision cosmology. The large volume of the resulting datasets, however, implies that their processing, classification, and interpretation will require sophisticated algorithms involving statistical learning. In this context, one important question to answer is: how do we discover the unexpected when we are presented with a large dataset? How do we find scientifically interesting light curves (or any kind of astronomical data) that are not explained by current models? In this talk I will discuss state-of-the-art anomaly detection methods that use machine learning to find needles in this upcoming haystack of data, and will show the results of applying them to a dataset of Kepler, TESS, and Chandra objects. After a brief introduction to machine learning and its application in time-domain astronomy, I will delve into different methods for outlier detection. I will then show how these methods can be adapted for time-domain and for high energy astronomy, and present the results of applying them to a large dataset of TESS light curves and the Chandra Source Catalog 2.0. I will describe the astrophysical implications of our findings in terms of where the most extreme outliers live in the Hertzprung-Russell diagram, and discuss the potential of the algorithms for discovery in the era of large astronomical datasets.

Mar 5th, 3pm, 131 Campbell

Deep learning has revolutionized big data—from outperforming doctors in skin cancer diagnoses to precisely forecasting earthquake aftershocks. In astronomy, deep discriminative models have been applied with great success to problems like galaxy classification and exoplanet identification. On the other hand, applications of powerful generative models are scarce. Here, we apply generative adversarial networks (GANs), a model composed of two dueling neural networks, to a variety of open problems in galaxy evolution and cosmology, namely: (1) deblending superpositions of distant galaxies to salvage galaxy images captured in the densest regions of the universe by near-future surveys like LSST, (2) super-resolving optical Suprime-Cam galaxy images from the COSMOS field to near Hubble quality to recover useful features for improved study of galaxy morphology and evolution, and (3) inferring the Lyman-alpha emission of high-redshift quasars given their redward spectrum to extract information about the early universe intergalactic medium.

Feb 5th, 4pm, 131 Campbell

Upcoming astronomical surveys such as the Large Synoptic Survey Telescope (LSST) will discover up to 10,000 new astronomical transients per night. However, LSST will not have enough resources to obtain the spectroscopy for each of these transients that is traditionally used to classify them. As a result, there has been a push to develop machine learning algorithms that can automatically classify transients using only photometric lightcurves. Transient classification has several major challenges for machine learning algorithms, including sparse measurements with heteroscedastic noise, highly unbalanced classes, and unrepresentative training samples. To address these issues, I developed an algorithm that models the lightcurves of transients using Gaussian process regression, and then classifies transients using tree based learning algorithms. This model took first place out of over 1000 entries in the recent LSST PLAsTiCC photometric classification challenge that was hosted on the Kaggle platform. In this talk, I will provide an overview of the LSST PLAsTiCC challenge and dataset, and I will describe the algorithm that I developed for transient classification.