Events

Open SkAI 2025

SkAI Institute Annual Conference

Dates: September 2–5, 2025

Location: SkAI Hub (172 E. Chestnut St., Suite 3500), Chicago, IL 60611

Conference website: link

Chicago skyline showing John Hancock Center

SkAI Institute events are held at the SkAI Hub in the John Hancock Center, Suite 3500, unless otherwise indicated, and are in Central (Standard) Time. If you have any questions about the events, please contact Emma Alexander (ealexander at northwestern.edu) and Alex Ćiprijanović (aleksand at fnal.gov) and, for logistics, please contact SkAI Communications and Events Manager, Gema Tinoco (gema.tinoco at northwestern.edu). For questions about a Journal Club meeting, please contact Xinfeng Xu (xinfeng.xu@northwestern.edu).

All SkAI events are presented in a hybrid (in-person and Zoom) format. Whether you plan to attend in person or virtually, please RSVP using the registration forms below, which will help us to get an accurate head count for catering and reporting purposes, and to make sure a visitor pass is requested for everyone who needs one. Please note that participants agree to follow the SkAI Institute Code of Conduct.

First time visiting the SkAI Hub? We have a brief orientation to help you know what to expect.

Location: John Hancock Center, Suite 3500 (unless otherwise indicated)
Zoom: Link

Registration Forms:
Registration Form for September 10 Works-in-Progress Wednesday

Upcoming Events:
Click on the speaker’s photo or name to learn more about them. Click on the title to reveal the abstract.

Date/Location	Speaker		Affiliation	Title	Links

Upcoming

September 10, 2025 Works-in-Progress Wednesday 10:00–11:00 a.m. Hancock, 3500		Vicky Kalogera Aggelos Katsaggelos	Northwestern University	TBA.	—
11:00–11:55 a.m.		Billy Moses Joaquin Vieira	University of Illinois Urbana-Champaign	TBA.	—
September 17, 2025 Journal Club 10:00–11:00 a.m. Hancock, 3500	TBA TBA	TBA TBA	TBA	TBA. TBA.	— —
October 1, 2025 Journal Club 10:00–11:00 a.m. Hancock, 3500	TBA TBA	TBA TBA	TBA	TBA. TBA.	— —
October 8, 2025 Works-in-Progress Wednesday 10:00–11:00 a.m. Hancock, 3500		Noelle Samia Gautham Narayan	Northwestern University University of Illinois Urbana-Champaign	TBA.	—
11:00–11:55 a.m.		Adam Miller Han Liu	Northwestern University	TBA.	—
October 15, 2025 Colloquium 10:00–11:55 a.m. Hancock, 3500	TBA	TBA	TBA	TBA.	—
October 22, 2025 Works-in-Progress Wednesday 10:00–11:00 a.m. Hancock, 3500		Tjitske Starkenburg Emma Alexander	Northwestern University	TBA.	—
11:00–11:55 a.m.		Ermin Wei Jeff McMahon	Northwestern University The University of Chicago	TBA.	—
November 5, 2025 Journal Club 10:00–11:00 a.m. Hancock, 3500	TBA TBA	TBA TBA	TBA	TBA. TBA.	— —
November 12, 2025 Colloquium 10:00–11:55 a.m. Hancock, 3500	TBA	TBA	TBA	TBA.	—
November 19, 2025 Works-in-Progress Wednesday 10:00–11:00 a.m. Hancock, 3500		Rina Barber Daniel Holz	The University of Chicago	TBA.	—
11:00–11:55 a.m.		Rebecca Willett Chihway Chang	The University of Chicago	TBA.	—
December 3, 2025 Journal Club 10:00–11:00 a.m. Hancock, 3500	TBA TBA	TBA TBA	TBA	TBA. TBA.	— —
December 17, 2025 Works-in-Progress Wednesday 10:00–11:00 a.m. Hancock, 3500		Aleksandra Ćiprijanović Sandeep Madireddy	Fermilab Argonne National Laboratory	TBA.	—
11:00–11:55 a.m.		Alex Drlica-Wagner Aravindan Vijayaraghavan Paul Chichura	The University of Chicago Northwestern University SkAI Institute	TBA.	—
January 14, 2026 Colloquium 10:00–11:55 a.m. Hancock, 3500	TBA	TBA	TBA	TBA.	—
January 21, 2026 Journal Club 10:00–11:00 a.m. Hancock, 3500	TBA TBA	TBA TBA	TBA	TBA. TBA.	— —
January 28, 2026 Works-in-Progress Wednesday 10:00–11:00 a.m. Hancock, 3500	TBA TBA	TBA TBA	TBA TBA	TBA.	—
11:00–11:55 a.m	TBA TBA	TBA TBA	TBA TBA	TBA.	—
February 4, 2026 Journal Club 10:00–11:55 a.m. Hancock, 3500	TBA TBA	TBA TBA	TBA	TBA. TBA.	— —
February 11, 2026 Works-in-Progress Wednesday 10:00–11:00 a.m. Hancock, 3500		Rina Barber Daniel Holz	The University of Chicago	TBA.	—
11:00–11:55 a.m	TBA TBA	TBA TBA	TBA TBA	TBA.	—
February 18, 2026 Colloquium 10:00–11:55 a.m. Hancock, 3500	TBA	TBA	TBA	TBA.	—
March 4, 2026 Journal Club 10:00–11:00 a.m. Hancock, 3500	TBA TBA	TBA TBA	TBA	TBA. TBA.	— —
March 11, 2026 Works-in-Progress Wednesday 10:00–11:00 a.m. Hancock, 3500		Adam Miller Han Liu	Northwestern University	TBA.	—
11:00–11:55 a.m		Ermin Wei Jeff McMahon	Northwestern University The University of Chicago	TBA.	—
March 18, 2026 Colloquium 10:00–11:55 a.m. Hancock, 3500	TBA	TBA	TBA	TBA.	—
April 1, 2026 Journal Club 10:00–11:00 a.m. Hancock, 3500	TBA TBA	TBA TBA	TBA	TBA. TBA.	— —
April 8, 2026 Works-in-Progress Wednesday 10:00–11:00 a.m. Hancock, 3500		Rebecca Willett Chihway Chang	The University of Chicago	TBA.	—
11:00–11:55 a.m		Noelle Samia Gautham Narayan	Northwestern University University of Illinois Urbana-Champaign	TBA.	—
April 15, 2026 Colloquium 10:00–11:55 a.m. Hancock, 3500	TBA	TBA	TBA	TBA.	—
April 22, 2026 Works-in-Progress Wednesday 10:00–11:00 a.m. Hancock, 3500		Aleksandra Ćiprijanović Sandeep Madireddy	Fermilab Argonne National Laboratory	TBA.	—
11:00–11:55 a.m		Tjitske Starkenburg Emma Alexander	Northwestern University	TBA.	—
May 6, 2026 Journal Club 10:00–11:00 a.m. Hancock, 3500	TBA TBA	TBA TBA	TBA	TBA. TBA.	— —
May 13, 2026 Colloquium 10:00–11:55 a.m. Hancock, 3500	TBA	TBA	TBA	TBA.	—

Past

August 29, 2025 Journal Club and Featured Talk 11:00–11:45 a.m. Hancock, 3500		Andreas Berlind	NSF Astronomy Division	TBA.	—
10:00–11:00 a.m.		Elizabeth Teng	Northwestern University	TBA.	—
May 28, 2025 Works-in-Progress Wednesday, SkAI Hub Orientation, and Journal Club 11:00–11:55 a.m. Hancock, 4010		Aldana Grichener Dalya Baron	The University of Arizona Stanford University	One of the main challenges in modeling massive stars to the onset of core collapse is the computational bottleneck of nucleosynthesis during advanced burning stages. The number of isotopes formed requires solving a large set of fully-coupled stiff ordinary differential equations (ODEs), making the simulations computationally intensive and prone to numerical instability. To overcome this barrier, we design a nuclear neural network (NNN) framework with multiple hidden layers to emulate nucleosynthesis calculations and conduct a proof-of-concept to evaluate its performance. The NNN takes the temperature, density and composition of a burning region as input and predicts the resulting isotopic abundances along with the energy generation and loss rates. We generate training sets for initial conditions corresponding to oxygen core depletion and beyond using large nuclear reaction networks, and compare the predictions of the NNNs to results from a commonly used small net. We find that the NNNs improve the accuracy of the electron fraction by 280–660% and the nuclear energy generation by 250–750%, consistently outperforming the small network across all timesteps. They also achieve significantly better predictions of neutrino losses on relatively short timescales, with improvements ranging from 100–10⁶%. While further work is needed to enhance their accuracy and applicability to different stellar conditions, integrating NNN trained models into stellar evolution codes is promising for facilitating large-scale generation of core-collapse supernova (CCSN) progenitors with higher physical fidelity. The structure and chemistry of the dusty interstellar medium (ISM) are shaped by complex processes that depend on the local radiation field, gas composition, and dust grain properties. Of particular importance are polycyclic aromatic hydrocarbons (PAHs), which emit strong vibrational bands in the mid-infrared, and play a key role in the ISM energy balance. We recently identified global correlations between PAH band and optical line ratios across three nearby galaxies, suggesting a connection between PAH heating and gas ionization throughout the ISM. In this work, we perform a census of the PAH heating–gas ionization connection using ∼700,000 independent pixels that probe scales of 40–150 pc in 19 nearby star-forming galaxies from the PHANGS survey. We find a universal relation between PAH(11.3 μm/7.7 μm) and ([S II]/Hα) with a slope of ∼0.2 and a scatter of ∼0.025 dex. The only exception is a group of anomalous pixels that show unusually high (11.3 μm/7.7 μm) PAH ratios in regions with old stellar populations and high starlight-to-dust emission ratios. Their mid-infrared spectra resemble those of elliptical galaxies. Active galactic nucleus hosts show modestly steeper slopes, with a ∼10% increase in PAH(11.3 μm/7.7 μm) in the diffuse gas on kiloparsec scales. This universal relation implies an emerging simplicity in the complex ISM, with a sequence that is driven by a single varying property: the spectral shape of the interstellar radiation field. This suggests that other properties, such as gas-phase abundances, gas ionization parameter, and grain charge distribution, are relatively uniform in all but specific cases.	Article Article
10:00–11:00 a.m.	SkAI Operations Team	Julian Cuevas-Zepeda	The University of Chicago Northwestern University	TBA. TBA.	— —
May 21, 2025 Works-in-Progress Wednesday and Colloquium 10:00–11:55 a.m. Hancock, 35th Floor		Weijian Li Souvik Chakraborty	Northwestern University Indian Institute of Technology (IIT) Delhi	TBA. Operator learning is an emerging area in scientific machine learning which aims to learn mappings between infinite dimensional function spaces. In the first half of the talk, I will delve into the foundations of Wavelet Neural Operator (WNO), a recently developed operator learning algorithm. I will discuss its working principles and its potential applications in complex engineering problems including fracture propagation in materials, tumor detection using USG data and elastography, and climate modelling. The second half of the talk will focus on what lies beyond neural operators. I will introduce a new scientific machine learning architecture that is loosely motivated from cognitive science. This architecture is a first of its kind foundation model and offers two key advantages: (i) it can simultaneously learn solution operators for multiple parametric PDEs, and (ii) rapid generalization to new parametric PDEs with minimal fine-tuning. We observe that the proposed architecture is robust against catastrophic forgetting and facilitate knowledge transfer across dissimilar tasks. Across a diverse array of mechanics problems, consistent performance enhancements are observed with this architecture compared to task-specific baseline operator learning frameworks.	— —
May 14, 2025 Journal Club 11:00–11:55 a.m. Hancock, 4010		Jason Sun Philipp Rajah Moura Srivastava	Northwestern University	The first measurements of the 21 cm brightness temperature power spectrum from the epoch of reionization will very likely be achieved in the near future by radio interferometric array experiments such as the Hydrogen Epoch of Reionization Array (HERA) and the Square Kilometre Array (SKA). Standard MCMC analyses use an explicit likelihood approximation to infer the reionization parameters from the 21 cm power spectrum. In this paper, we present a new Bayesian inference of the reionization parameters where the likelihood is implicitly defined through forward simulations using density estimation likelihood-free inference (DELFI). Realistic effects, including thermal noise and foreground avoidance, are also applied to the mock observations from the HERA and SKA. We demonstrate that this method recovers accurate posterior distributions for the reionization parameters, and it outperforms the standard MCMC analysis in terms of the location and size of credible parameter regions. With the minute-level processing time once the network is trained, this technique is a promising approach for the scientific interpretation of future 21 cm power spectrum observation data. Our code 21cmDELFI-PS is publicly available at this link (https://github.com/Xiaosheng-Zhao/21cmDELFI). Tomographic three-dimensional 21 cm images from the epoch of reionization contain a wealth of information about the reionization of the intergalactic medium by astrophysical sources. Conventional power spectrum analysis cannot exploit the full information in the 21 cm data because the 21 cm signal is highly non-Gaussian due to reionization patchiness. We perform a Bayesian inference of the reionization parameters where the likelihood is implicitly defined through forward simulations using density estimation likelihood-free inference (DELFI). We adopt a trained 3D convolutional neural network (CNN) to compress the 3D image data into informative summaries (DELFI-3D CNN). We show that this method recovers accurate posterior distributions for the reionization parameters. Our approach outperforms earlier analysis based on two-dimensional 21 cm images. In contrast, a Monte Carlo Markov Chain analysis of the 3D light-cone-based 21 cm power spectrum alone and using a standard explicit likelihood approximation results in less accurate credible parameter regions than inferred by the DELFI-3D CNN, both in terms of the location and shape of the contours. Our proof-of-concept study implies that the DELFI-3D CNN can effectively exploit more information in the 3D 21 cm images than a 2D CNN or power spectrum analysis. This technique can be readily extended to include realistic effects and is therefore a promising approach for the scientific interpretation of future 21 cm observation data. Modeling of large populations of binary stellar systems is an integral part of many areas of astrophysics, from radio pulsars and supernovae to X-ray binaries, gamma-ray bursts, and gravitational-wave mergers. Binary population synthesis codes that employ self-consistently the most advanced physics treatment available for stellar interiors and their evolution and are at the same time computationally tractable have started to emerge only recently. One element that is still missing from these codes is the ability to generate the complete time evolution of binaries with arbitrary initial conditions using precomputed three-dimensional grids of binary sequences. Here, we present a highly interpretable method, from binary evolution track interpolation. Our method implements simulation generation from irregularly sampled time series. Our results indicate that this method is appropriate for applications within binary population synthesis and computational astrophysics with time-dependent simulations in general. Furthermore, we point out and offer solutions to the difficulty surrounding evaluating the performance of signals exhibiting extreme morphologies akin to discontinuities.	Article Article Article
May 7, 2025 Works-in-Progress Wednesday 11:20–11:55 a.m. Hancock, 35th Floor		Ugur Demir	Northwestern University	TBA.	—
10:40–11:20 a.m.		Roxie (Ruoxi) Jiang	The University of Chicago	TBA.	—
10:00–10:40 a.m.		Tri Nguyen	Northwestern University	TBA.	—
April 30, 2025 Journal Club 11:00–11:55 a.m. Hancock, 4010		Elizabeth Teng Kyle Rocha	Northwestern University	Knowledge about the internal physical structure of stars is crucial to understanding their evolution. The novel binary population synthesis code POSYDON includes a module for interpolating the stellar and binary properties of any system at the end of binary MESA evolution based on a pre-computed set of models. In this work, we present a new emulation method for predicting stellar profiles, i.e., the internal stellar structure along the radial axis, using machine learning techniques. We use principal component analysis for dimensionality reduction and fully-connected feed-forward neural networks for making predictions. We find accuracy to be comparable to that of nearest neighbor approximation, with a strong advantage in terms of memory and storage efficiency. By providing a versatile framework for modeling stellar internal structure, the emulation method presented here will enable faster simulations of higher physical fidelity, offering a foundation for a wide range of large-scale population studies of stellar and binary evolution. Dusty stellar point sources are a significant stage in stellar evolution and contribute to the metal enrichment of galaxies. These objects can be classified using photometric and spectroscopic observations with color-magnitude diagrams (CMD) and infrared excesses in spectral energy distributions (SED). We employed supervised machine learning spectral classification to categorize dusty stellar sources, including young stellar objects (YSOs) and evolved stars (oxygen- and carbon-rich asymptotic giant branch stars, AGBs), red supergiants (RSGs), and post-AGB (PAGB) stars in the Large and Small Magellanic Clouds, based on spectroscopic labeled data from the Surveying the Agents of Galaxy Evolution (SAGE) project, which used 12 multiwavelength filters and 618 stellar objects. Despite missing values and uncertainties in the SAGE spectral datasets, we achieved accurate classifications. To address small and imbalanced spectral catalogs, we used the Synthetic Minority Oversampling Technique (SMOTE) to generate synthetic data points. Among models applied before and after data augmentation, the Probabilistic Random Forest (PRF), a tuned Random Forest (RF), achieved the highest total accuracy, reaching 89% based on recall in categorizing dusty stellar sources. Using SMOTE does not improve the best model's accuracy for the CAGB, PAGB, and RSG classes; it remains 100%, 100%, and 88%, respectively, but shows variations for OAGB and YSO classes. We also collected photometric labeled data similar to the training dataset, classifying them using the top four PRF models with over 87% accuracy. Multiwavelength data from several studies were classified using a consensus model integrating four top models to present common labels as final predictions.	Article Article
April 23, 2025 Works-in-Progress Wednesday 11:20–11:55 a.m. Hancock, 4010		Sneh Pandya	Northeastern University	TBA.	—
10:40–11:20 a.m.		Jennifer Li	University of Illinois Urbana-Champaign	TBA.	—
10:00–10:40 a.m.		Anarya Ray	Northwestern University	TBA.	—
April 16, 2025 Journal Club 11:00–11:55 a.m. Hancock, 4010		Anowar J. Shajib Tri Nguyen	The University of Chicago Northwestern University	Strong gravitational lensing is a powerful tool for probing the internal structure and evolution of galaxies, the nature of dark matter, and the expansion history of the Universe, among many other scientific applications. For almost all of these science cases, modeling the lensing mass distribution is essential. For that, forward modeling of imaging data to the pixel level is the standard method used for galaxy-scale lenses. However, the traditional workflow of forward lens modeling necessitates a significant amount of human investigator time, requiring iterative tweaking and tuning of the model settings through trial and error. An automated lens modeling pipeline can substantially reduce the need for human investigator time. In this paper, we present \textsc{dolphin}, an automated lens modeling pipeline that combines artificial intelligence with the traditional forward modeling framework to enable full automation of the modeling workflow. \textsc{dolphin} uses a neural network model to perform visual recognition of the strong lens components, then autonomously sets up a lens model with appropriate complexity, and fits the model with the modeling engine, lenstronomy. Thanks to the versatility of lenstronomy, dolphin can autonomously model both galaxy-galaxy and galaxy-quasar strong lenses. The phase space of stellar streams is proposed to detect dark substructure in the Milky Way through the perturbations created by passing subhalos - and thus is a powerful test of the cold dark matter paradigm and its alternatives. Using graph convolutional neural network (GCNN) data compression and simulation-based inference (SBI) on a simulated GD-1-like stream, we improve the constraint on the mass of a [10⁸, 10⁷, 10⁶] M_⊙ perturbing subhalo by factors of [11, 7, 3] with respect to the current state-of-the-art density power spectrum analysis. We find that the GCNN produces posteriors that are more accurate (better calibrated) than the power spectrum. We simulate the positions and velocities of stars in a GD-1-like stream and perturb the stream with subhalos of varying mass and velocity. Leveraging the feature encoding of the GCNN to compress the input phase space data, we then use SBI to estimate the joint posterior of the subhalo mass and velocity. We investigate how our results scale with the size of the GCNN, the coordinate system of the input and the effect of incomplete observations. Our results suggest that a survey with 10× fewer stars (300 stars) with complete 6-D phase space data performs about as well as a deeper survey (3000 stars) with only 3-D data (photometry, spectroscopy). The stronger constraining power and more accurate posterior estimation motivate further development of GCNNs in combining future photometric, spectroscopic and astrometric stream observations.	Article Article
April 9, 2025 Works-in-Progress Wednesday 11:00–11:55 a.m. Hancock, 4010		Aleksandra Ćiprijanović Sandeep Madireddy	Fermilab Argonne National Laboratory	TBA.	—
10:00–11:00 a.m.		Rebecca Willett Chihway Chang	The University of Chicago	TBA.	—
April 4, 2025 Colloquium 10:30–11:55 a.m. Hancock, 4010		Alyssa Goodman	Harvard University	I have been lucky enough in my career so far to watch, and I hope help, computational technology change what we can learn about our Universe. Today, in 2025, I somewhat unexpectedly find myself involved in a broad range of AI-based and AI-enhanced efforts designed to speed learning and discovery in astrophysics and in science. In this talk, I will offer glimpses into a handful of ongoing AI-enhanced efforts, each of which is very different from the others, yet which work together in a researcher/educator's life to speed progress. Work to be highlighted includes: (1) automated data-set linking in the “glue” and LIVE-Environments visualization environments; (2) The “Reading Time Machine,” which uses AI to “read” graphics and images and ingest their content into the ADS Literature archive, as “data”; (3) approaches to 3D selection in volumetric data, using both AI and augmented reality (AR); (4) a quest to understand why LLMs are so good at describing infographics, but so terrible at creating them; (5) capabilities of AI for writing code for visualization, in both research and education. The plan of the talk will be to present an overview of each of these efforts in order to inspire broader discussion of whichever topics evolve as most interesting to the assembled audience.	—
April 02, 2025 Journal Club 11:00–11:55 a.m. Hancock, 4010		George Winstone Cliff Johnson	Northwestern University	Gravitational waves, detected a century after they were first theorized, are spacetime distortions caused by some of the most cataclysmic events in the universe, including black hole mergers and supernovae. The successful detection of these waves has been made possible by ingenious detectors designed by human experts. Beyond these successful designs, the vast space of experimental configurations remains largely unexplored, offering an exciting territory potentially rich in innovative and unconventional detection strategies. Here, we demonstrate the application of artificial intelligence (AI) to systematically explore this enormous space, revealing novel topologies for gravitational wave (GW) detectors that outperform current next-generation designs under realistic experimental constraints. Our results span a broad range of astrophysical targets, such as black hole and neutron star mergers, supernovae, and primordial GW sources. Moreover, we are able to conceptualize the initially unorthodox discovered designs, emphasizing the potential of using AI algorithms not only in discovering but also in understanding these novel topologies. We've assembled more than 50 superior solutions in a publicly available Gravitational Wave Detector Zoo which could lead to many new surprising techniques. At a bigger picture, our approach is not limited to gravitational wave detectors and can be extended to AI-driven design of experiments across diverse domains of fundamental physics. We present a catalogue of 497 galaxy-galaxy strong lenses in the Euclid Quick Release 1 data (63 deg²). In the initial 0.45\% of Euclid's surveys, we double the total number of known lens candidates with space-based imaging. Our catalogue includes 250 grade A candidates, the vast majority of which (243) were previously unpublished. Euclid's resolution reveals rare lens configurations of scientific value including double-source-plane lenses, edge-on lenses, complete Einstein rings, and quadruply-imaged lenses. We resolve lenses with small Einstein radii ($\theta_{\rm E} < \ang{;;1}$) in large numbers for the first time. These lenses are found through an initial sweep by deep learning models, followed by Space Warps citizen scientist inspection, expert vetting, and system-by-system modelling. Our search approach scales straightforwardly to Euclid Data Release 1 and, without changes, would yield approximately 7000 high-confidence (grade A or B) lens candidates by late 2026. Further extrapolating to the complete Euclid Wide Survey implies a likely yield of over 100000 high-confidence candidates, transforming strong lensing science.	Article Article
March 26, 2025 Works-in-Progress Wednesday 11:20–11:55 a.m. Hancock, 4010		Alex Drlica-Wagner Aravindan Vijayaraghavan	The University of Chicago Northwestern University	TBA.	—
10:00–10:50 a.m.		Rina Barber Daniel Holz	The University of Chicago	TBA.	—
March 19, 2025 Journal Club 11:00–11:55 a.m. Hancock, 4010		Tianao Li Xinfeng Xu	Northwestern University	Diffusion models recently proved to be remarkable priors for Bayesian inverse problems. However, training these models typically requires access to large amounts of clean data, which could prove difficult in some settings. In this work, we present a novel method based on the expectation-maximization algorithm for training diffusion models from incomplete and noisy observations only. Unlike previous works, our method leads to proper diffusion models, which is crucial for downstream tasks. As part of our method, we propose and motivate an improved posterior sampling scheme for unconditional diffusion models. We present empirical evidence supporting the effectiveness of our method. Mergers of binary neutron stars emit signals in both the gravitational-wave (GW) and electromagnetic spectra. Famously, the 2017 multi-messenger observation of GW170817 led to scientific discoveries across cosmology, nuclear physics and gravity. Central to these results were the sky localization and distance obtained from the GW data, which, in the case of GW170817, helped to identify the associated electromagnetic transient, AT 2017gfo, 11 h after the GW signal. Fast analysis of GW data is critical for directing time-sensitive electromagnetic observations. However, owing to challenges arising from the length and complexity of signals, it is often necessary to make approximations that sacrifice accuracy. Here we present a machine-learning framework that performs complete binary neutron star inference in just 1 s without making any such approximations. Our approach enhances multi-messenger observations by providing: (1) accurate localization even before the merger; (2) improved localization precision by around 30% compared to approximate low-latency methods; and (3) detailed information on luminosity distance, inclination and masses, which can be used to prioritize expensive telescope time. Additionally, the flexibility and reduced cost of our method open new opportunities for equation-of-state studies. Finally, we demonstrate that our method scales to long signals, up to an hour in length, thus serving as a blueprint for data analysis for next-generation ground- and space-based detectors.	Article Article
March 12, 2025 Colloquium 10:30–11:30 a.m. Hancock, 4010		Bill Gropp	University of Illinois Urbana-Champaign	The end of Dennard or frequency scaling in computer processors nearly twenty years ago has caused a transformation in computing. Innovations in computer architecture have enabled continued improvements in performance, but at the cost of increasing software complexity. GPUs have been key in providing performance for many applications and have enabled the revolution in machine learning and AI. This talk will provide some background on the transformations in computing over the last two decades, describe NCSA and its approach to this revolution in computing, and close with a description of NCSA's efforts in AI.	—
March 3, 2025 Colloquium 11:00–11:55 a.m. Hancock, 4010		Renée Hložek	University of Toronto	In the sky: the Legacy Survey of Space and Time (LSST) on the Vera C. Rubin Observatory will generate a data deluge: millions of astronomical transients and variable sources will need to be classified from their light curves. I'll discuss the efforts within the Dark Energy Science Collaboration (DESC) to get ready for transient classification through efforts like public Photometric LSST Astronomical Time-series Classification Challenge (PLAsTiCC) and the Extended LSST Astronomical Time-series Classification Challenge (ELAsTiCC) was an expert challenge to LSST broker teams themselves to classify alert streams. I'll place this work in the context of pushing from detections to cosmology. Looking to the brain: I'll present AstroBEATS, a pipeline derived from astronomical image analysis techniques and designed for high-resolution images of the brain. I'll describe how AstroBEATS can be used to study the synaptic firing in the brain and to search for signs of neurodegeneration and describe the processes that generated this interdisciplinary research.	—
February 12, 2025 Works-in-Progress Wednesday 11:20–11:55 a.m. Hancock, 4010		Jeff McMahon Ermin Wei	The University of Chicago Northwestern University	TBA.	—
10:45–11:20 a.m.		Vicky Kalogera Aggelos Katsaggelos	Northwestern University	TBA.	—
10:00–10:45 a.m.		Emma Alexander Tjitske Starkenburg	Northwestern University	TBA.	—

The SkAI Institute is one of the National Artificial Intelligence Research Institutes funded by the U.S. National Science Foundation and Simons Foundation. Information on National AI Institutes is available at aiinstitutes.org.

Mailing address:
SkAI Institute
875 N. Michigan Ave., Suite 3500
Chicago, IL 60611
skai-institute at u.northwestern.edu
(Get to the SkAI Hub via the 172 E. Chestnut St. entrance)