• Machine-Learning Accelerated Calculations of Reduced Density Matrices: arXiv:2511.07367  

Strongly correlated quantum systems are famously hard to compute, even for experts with powerful computers, because tracking how many particles interact quickly becomes intractable in a high-dimensional Hilbert space. In this work, we build two AI architectures that learn from small systems and predict the physics of much larger systmes: one self-attention–based neural network (inspired by Transformers) takes rough, unphysical guesses of n-particle reduced density matrices (n-RDMs) and “repairs” them into valid physical ones, while another (a SIREN network) learns the smooth structure of these correlations across momentum space and then interpolates them to larger lattices than it ever saw in training. 

Across several benchmark physical models, these AIs not only reproduce detailed many-body correlation patterns but also dramatically accelerate traditional methods like Hartree–Fock. Together, they suggest a future in which AI becomes an assistant for theorists. More precisely,  our results illustrate the potential of using NN-based methods for interpolable physical quantities, which might open a new avenue for future research on strongly correlated phases.

• THE ASTROPHYSICAL JOURNAL:  https://iopscience.iop.org/article/10.3847/1538-4357/adfb7d 

Our research (https://www.arxiv.org/abs/2508.09370) delivers a practical advance for astrophysics: a ML approach that reliably classifies the coolest stars and brown dwarfs—objects that dominate our Galaxy and often host planets—without slow, expert-by-eye inspection. Using low-resolution near-infrared spectra, our models achieve ~95% accuracy within one spectral subtype and can also flag key physical traits (surface gravity and metallicity), a capability that scales to the millions of spectra expected from surveys like Gaia, SDSS, and SPHEREx. By automating and standardizing this step, the work accelerates cataloging nearby low-mass objects, improves the search for unusual targets, and strengthens population studies that inform planet formation and Galactic structure.

• American Astronomical Society: https://ui.adsabs.harvard.edu/abs/2024AAS...24333006Z/abstract 

We analyzed 219 million low-resolution Gaia DR3 spectra and, after rigorous quality cuts, distilled a high-purity set of 37,945 stellar spectra to build calibrated templates. Anchored to SDSS spectral standards, the library spans 3,360–10,200 Å and includes 201 main-sequence templates across OBAFGKM (subtypes 0–9), plus white dwarfs (DA0.5–DA7.0) and carbon stars (dCG, dCK, dCM). The result is a clean, standardized reference for fast, consistent spectral typing and downstream stellar population studies. Github Repo 

• American Astronomical Society:  https://ui.adsabs.harvard.edu/abs/2025AAS...24546403Z/abstract 

We build ML tools that read low-resolution infrared spectra to identify the coolest stars and brown dwarfs (M, L, T) with about 95% accuracy—no hand-typing required. Beyond labels, the models also flag metallicity and surface gravity clues, helping us quickly map and understand our smallest, coolest galactic neighbors.