Presentations 

Hyungsuk Tak (PSU) Sep 4, 2024

 Six Maxims of Statistical Acumen for Astronomical Data Analysis
 Abstract: The production of complex astronomical data is accelerating, especially with newer telescopes producing ever more largescale surveys. The increased quantity, complexity, and variety of astronomical data demand a parallel increase in skill and sophistication in developing, deciding, and deploying statistical methods. Understanding limitations and appreciating nuances in statistical and machine learning methods and the reasoning behind them is essential for improving dataanalytic proficiency and acumen. Aiming to facilitate such improvement in astronomy, we delineate cautionary tales in statistics via six maxims, with examples drawn from the astronomical literature. Inspired by the significant quality improvement in business and manufacturing processes by the routine adoption of Six Sigma, we hope the routine reflection on these Six Maxims will improve the quality of both data analysis andscientific findings in astronomy.
 See also: arXiv:2409.16179 [!arXiv]
 Presentation slides [.pdf]
 Presentation video [!yt]


Yanke Song (Harvard) Sep 18, 2024

 A Poissonprocess AutoDecoder for Xray Sources
 Abstract: Xray observing facilities such as the Chandra Xray Observatory and the eROSITA all sky survey have detected millions of astronomical sources associated with highenergy phenomena. The arrival of photons as a function of time follows a Poisson process and can vary by ordersofmagnitude, presenting obstacles for downstream tasks such as source classification, physical property derivation and anomaly detection. Previous work has either failed to directly capture the Poisson nature of the data or only focuses on Poisson rate function reconstruction. In this work, we present Poisson Process AutoDecoder (PPAD). PPAD is a neural field decoder that maps fixedlength latent features to continuous Poisson rate functions across energy band and time via unsupervised learning. It reconstructs the rate function and yields a representation at the same time. We demonstrate the efficacy of PPAD reconstruction, regression, classification and anomaly detection experiments using the Chandra Source Catalog.
 Slides [.pdf] (slightly evolved from presented slides)
 Presentation video [!yt]


Ann Lee (CMU) Oct 16, 2024

 Valid Scientific Inference with Neural Density Estimators
 Abstract: Scientific inference often involves inferring internal key parameters that determine the outcome of a complex physical phenomenon. The data themselves may come in the form of a labeled set that implicitly encodes the likelihood function; for example, in the form of (i) pairs of internal parameters and observable data according to a mechanistic (simulator) model, or as (ii) observed data and internal parameters, where the latter parameters are not directly observable, but have been inferred to high precision via an auxiliary measurement and a theoretical model. We refer to inference in both of these intractable likelihood settings as "LikelihoodFree Inference'" (LFI). Scientists are increasingly leveraging machine learning methods, such as neural density estimator and AI generative models, for parameter inference in LFI settings. However, highposterior density regions derived from these density estimators do not necessarily have a high probability of including the true parameter of interest, even if the posterior is wellestimated and the labeled data have the same distribution as the target distribution. Furthermore, if the prior distribution is poorly specified, then the HPD regions could severely undercover and/or be biased, thereby leading to misleading scientific conclusions. In this talk, I will present new LFI methodology and algorithms for leveraging neural density estimators to produce confidence regions of parameters of interest that have (i) nominal frequentist coverage for any values of the true unknown parameters, and (ii) smaller average area (yielding higher constraining power) if the prior is wellspecified. I will illustrate our methods on examples from astronomy and highenergy physics, and discuss where we stand and what challenges still remain. (This work is joint with Luca Masserano, James Carzon, Antonio Carlos Herling Ribeiro Junior, Alex Shen, Tommaso Dorigo, Michele Doro, Mikael Kuusela, Joshua Speagle, Rafael Izbicki)
 Presentation slides [.pdf]
 Presentation Video [!yt]


Christina Reissel (MIT) Oct 30, 2024

 Machine Learning for RealTime Analysis of Gravitational Wave Data
 Abstract:
The detection of gravitational waves has opened a new field of astronomy. Upgrades to gravitational wave observatories will further increase the sensitivity to a broader range of astrophysical events, specifically to weaker signals from more distant or lowermass events, simultaneously increasing backgrounds dramatically. At the same time, artificial intelligence (AI) enables fast decisionmaking and interpretation with performances that transcend other strategies.
In this talk, we present ml4gw, an endtoend software framework for realtime gravitationalwave data analysis. Combining AIbased algorithms for denoising, binary black detection, anomaly detection, and realtime parameter estimation, we aim for a fast, intuitive pipeline which is able to leverage the powerful modeling techniques available in the gravitational wave literature. This realtime gravitationalwave data analysis will be an essential backbone for optimal multimessenger astronomy, providing fast and accurate triggers about potentially interesting events, enabling minimal reaction times of other observatories.
 Presentation Video [!yt]


Justina Yang (Harvard) Nov 6, 2024 Noon EST Zoom+SciCen 706

 Emulating Photon Pileup Effects on Xray Spectra with a Neural Network
 Abstract: In Xray astronomy, photon pileup occurs when two or more Xray photons strike a detector within the same readout time and are recorded as a single "piledup" event. In instruments such as the ACIS detector on the Chandra Xray Observatory, pileup can cause effects such as artificially low event count rates or the misidentification of events as events of higher energies. These effects distort the shape and intensity of observed spectra, especially for bright sources, and complicate the inference of source properties. However, pileup effects are challenging to model analytically and the standard statistical method for modeling pileup (Davis 2001) is not wellsuited to analyzing data with severe amounts of pileup. We therefore construct a neural network emulator to mimic the empirical effects of pileup on Xray spectra. We first limit the emulator to data simulated with MARX (Davis et al. 2012), examine the emulator's ability to mimic an approximate version of pileup, and describe how the emulator can be incorporated into standard parameter inference methods. We also discuss next steps for refining the neural network to emulate the empirical version of pileup in the ACIS detector. We expect that our method can be adapted for future Xray detectors to emulate instrumental effects that are analogous to pileup.


Robert Zimmerman (Toronto+Imperial) Nov 13, 2024 Zoom 
 Hidden Markov Modeling of Xray light curves


Herman Marshall (MIT) Nov 20, 2024 Noon EST SciCen 706 + Zoom 
 How can we test the origins of extragalactic neutrinos?
 Abstract: The processes that can produce high energy neutrinos are relevant to high energy physics and astrophysics, representing environments that cannot be produced on Earth. They are detected by a large experiment buried in the ice at the south pole. The experiment uses the bulk of the earth as a way to block background. Neutrinos have such a low cross section for interaction that they travel through the earth, interacting in the ice with upwardgoing tracks of secondary particles. They are extremely rare and have large error regions on the sky, so it is difficult to identify any specific source. There have been a few population studies using crosscorrelation methods, such as Buson et a. (2022, 2022ApJ...933L..43B). There are, of course, complications but Buson et al claim a pvalue of 6e7, seemingly irrefutable. After expressing interest (perhaps "doubt") in their methods, I was asked by Dr. Buson at a recent meeting to see if a set of statisticians could review their methods for validating the associations of neutrinos with radio loud quasars.


Matteo Guardini (ESO/Garching) Feb 22, 2025 Zoom 
 Automatic astrophysical component detection and modeling


Aritra Banerjee (Minnesota) Apr 2, 2025 
 Detecting signals under background mismodeling





