arXiv subjects

Jeff Shen

Publications and source records attributed to Jeff Shen.

18 recordsLinked to original sources

arXiv2026

Emergent Transfer of a Physics Foundation Model from Simulation to Laboratory Turbulence

Whether physics foundation models can be usefully deployed on laboratory experiments remains an open question for scientific machine learning (ML). We test this question on the Rayleigh-Taylor instability (RTI), a ubiquitous and demanding fluid instability seen from tabletop flows to supernova explosions, in which small perturbations at a density interface grow into chaotic, multiscale mixing as a lighter fluid accelerates into a heavier one. Standard ML models struggle with RTI, and despite over a century of theoretical, numerical, and experimental work, it carries an unresolved discrepancy between simulation and experiment: the late-time mixing growth rate, $\alpha$, measured in most laboratory experiments ($\sim$ 0.06-0.07), is roughly three times the value from idealized direct numerical simulations (DNS, $\sim$ 0.02). The gap's origin remains debated. These properties make RTI a stringent test for a question that matters well beyond RTI: can foundation models trained only on simulations generalise to sparse, messy, and noisy laboratory settings? We finetune Walrus, a foundation model for continuum dynamics, on three or fewer DNS realizations and recover key RTI physics over long rollouts. Applied zero-shot to sliding-barrier laboratory data, the finetuned model leaves the DNS-like regime and enters the observed growth band, having never seen a single experimental sample. These results provide independent, data-driven evidence that initial conditions play a crucial role in the longstanding sim-experiment gap in $\alpha$. The model also generalises zero-shot to stable stratification, a buoyancy regime absent from training, correctly slowing mixing-layer growth. Together, our results show that foundation models can generalise well beyond their training data, predicting laboratory behavior and unseen physical regimes, opening new ways to probe longstanding simulation-experiment gaps.

physics.flu-dyn↗

arXiv2026

Homogeneous Stellar Parameters from Heterogeneous Spectra with Deep Learning

Large-scale spectroscopic surveys have collectively observed millions of stars across the Milky Way, but each derives stellar labels using independent pipelines with distinct modelling assumptions, introducing systematic offsets that obscure signals in chemical space and hinder large-scale Galactic archaeology. We present a unified deep-learning framework that delivers atmospheric parameters, chemical abundances for 20 elements, distances, and ages -- all on a single, self-consistent scale -- for an arbitrary number of spectroscopic surveys simultaneously. Our approach uses a Transformer model that ingests spectra of arbitrary wavelength range and resolution, trained end-to-end as a single model across all surveys, eliminating the need for post-hoc recalibration. We apply this framework to spectra from APOGEE DR17, GALAH DR3, DESI DR1, and $\textit{Gaia}$ RVS DR3, spanning resolutions from R ~ 2,000 to 28,000 and wavelengths from the optical to the near-infrared. On high-resolution APOGEE spectra the model achieves precisions of $18~$K in $\textrm{T}_{\rm eff}$, $0.04~$dex in $\textrm{log}\,\textit{g}$, $0.015~$dex in [Fe/H], and ${<}\,0.03~$dex across all abundances; on lower-resolution DESI spectra, typical precisions are $51~$K, $0.09~$dex, $0.04~$dex, and ${\sim}\,0.06~$dex, respectively. Cross-survey comparisons demonstrate that labels for the same stars observed by different surveys are consistent within model uncertainties; we further validate against external distance catalogs and open cluster metallicities and ages. The resulting homogeneous catalog enables Galactic archaeology at unprecedented scale and consistency, and the framework is readily extensible to forthcoming spectroscopic surveys such as SDSS-V, WEAVE, and 4MOST. The catalog is publicly available at https://doi.org/10.5281/zenodo.19830515.

astro-ph.GA↗

arXiv2026

MIMIC: A Generative Multimodal Foundation Model for Biomolecules

Biological function emerges from coupled constraints across sequence, structure, regulation, evolution, and cellular context, yet most foundation models in biology are trained within one modality or for a fixed forward task. We present MIMIC, a generative multimodal foundation model trained on our newly curated and aligned dataset, LORE, linking nucleic acid, protein, evolutionary, structural, regulatory, and semantic/contextual modalities within partially observed biomolecular states. MIMIC uses a split-track encoder-decoder architecture to condition on arbitrary subsets of observed modalities and reconstruct or generate missing components of molecular state across the genome, transcriptome, and proteome. Multimodal conditioning consistently improves MIMIC's sequence reconstruction relative to sequence-only inputs, while its learned representations enable state-of-the-art performance on RNA and protein downstream tasks. MIMIC achieves state-of-the-art splicing prediction, and its joint generative formulation enables isoform-aware inference that further improves performance. Beyond prediction, the same generative framework supports constrained design. For RNA, MIMIC identifies corrective edits in a clinically relevant HBB splice-disrupting mutation without reverting it by using evolutionary and structural signals. For proteins, jointly conditioning on shape and surface chemistry of PD-L1 and hACE2 binding sites produces diverse, high-confidence sequences with strong in silico support for target binding. Finally, MIMIC uses experimental context as semantic conditioning to model assay-dependent RNA chemical probing, rather than treating context as a fixed output. Together, these results position MIMIC's aligned multimodal generative modeling as a strong foundation for unifying representation learning, conditional prediction, and constrained biomolecular design within a single model.

cs.AI↗

arXiv2025

Predicting partially observable dynamical systems via diffusion models with a multiscale inference scheme

Conditional diffusion models provide a natural framework for probabilistic prediction of dynamical systems and have been successfully applied to fluid dynamics and weather prediction. However, in many settings, the available information at a given time represents only a small fraction of what is needed to predict future states, either due to measurement uncertainty or because only a small fraction of the state can be observed. This is true for example in solar physics, where we can observe the Sun's surface and atmosphere, but its evolution is driven by internal processes for which we lack direct measurements. In this paper, we tackle the probabilistic prediction of partially observable, long-memory dynamical systems, with applications to solar dynamics and the evolution of active regions. We show that standard inference schemes, such as autoregressive rollouts, fail to capture long-range dependencies in the data, largely because they do not integrate past information effectively. To overcome this, we propose a multiscale inference scheme for diffusion models, tailored to physical processes. Our method generates trajectories that are temporally fine-grained near the present and coarser as we move farther away, which enables capturing long-range temporal dependencies without increasing computational cost. When integrated into a diffusion model, we show that our inference scheme significantly reduces the bias of the predicted distributions and improves rollout stability.

cs.LG↗

arXiv2025

Walrus: A Cross-Domain Foundation Model for Continuum Dynamics

Foundation models have transformed machine learning for language and vision, but achieving comparable impact in physical simulation remains a challenge. Data heterogeneity and unstable long-term dynamics inhibit learning from sufficiently diverse dynamics, while varying resolutions and dimensionalities challenge efficient training on modern hardware. Through empirical and theoretical analysis, we incorporate new approaches to mitigate these obstacles, including a harmonic-analysis-based stabilization method, load-balanced distributed 2D and 3D training strategies, and compute-adaptive tokenization. Using these tools, we develop Walrus, a transformer-based foundation model developed primarily for fluid-like continuum dynamics. Walrus is pretrained on nineteen diverse scenarios spanning astrophysics, geoscience, rheology, plasma physics, acoustics, and classical fluids. Experiments show that Walrus outperforms prior foundation models on both short and long term prediction horizons on downstream tasks and across the breadth of pretraining data, while ablation studies confirm the value of our contributions to forecast stability, training throughput, and transfer performance over conventional approaches. Code and weights are released for community use.

cs.LG↗

arXiv2025

Universal Spectral Tokenization via Self-Supervised Panchromatic Representation Learning

Sequential scientific data span many resolutions and domains, and unifying them into a common representation is a key step toward developing foundation models for the sciences. Astronomical spectra exemplify this challenge: massive surveys have collected millions of spectra across a wide range of wavelengths and resolutions, yet analyses remain fragmented across spectral domains (e.g., optical vs. infrared) and object types (e.g., stars vs. galaxies), limiting the ability to pool information across datasets. We present a deep learning model that jointly learns from heterogeneous spectra in a self-supervised manner. Our universal spectral tokenizer processes spectra from a variety of object types and resolutions directly on their native wavelength grids, producing intrinsically aligned, homogeneous, and physically meaningful representations that can be efficiently adapted to achieve competitive performance across a range of downstream tasks. For the first time, we demonstrate that a single model can unify spectral data across resolutions and domains, suggesting that our model can serve as a powerful building block for foundation models in astronomy -- and potentially extend to other scientific domains with heterogeneous sequential data, such as climate and healthcare.

astro-ph.IM↗

arXiv2025

AION-1: Omnimodal Foundation Model for Astronomical Sciences

While foundation models have shown promise across a variety of fields, astronomy still lacks a unified framework for joint modeling across its highly diverse data modalities. In this paper, we present AION-1, a family of large-scale multimodal foundation models for astronomy. AION-1 integrates heterogeneous imaging, spectroscopic, and scalar data using a two-stage architecture: modality-specific tokenization followed by transformer-based masked modeling of cross-modal token sequences. The model is pretrained on five large-scale surveys: Legacy Survey, Hyper Suprime-Cam (HSC), Sloan Digital Sky Survey (SDSS), Dark Energy Spectroscopic Instrument (DESI), and Gaia. These span more than 200 million observations of stars, galaxies, and quasars. With a single frozen encoder, AION-1 achieves strong results on a broad suite of downstream tasks, including galaxy and stellar property estimation, galaxy morphology classification, similarity-based retrieval, galaxy image segmentation, and spectral super-resolution. We release AION-1 model variants ranging from 300 M to 3.1 B parameters. Beyond astronomy, AION-1 provides a scalable blueprint for multimodal scientific foundation models that can seamlessly integrate noisy, instrument-specific observations. All code, tokenizers, pretrained weights, and a lightweight evaluation suite are released under an open-source license.

astro-ph.IM↗

arXiv2025

ALICE: An Interpretable Neural Architecture for Generalization in Substitution Ciphers

We present cryptogram solving as an ideal testbed for studying neural network reasoning and generalization; models must decrypt text encoded with substitution ciphers, choosing from 26! possible mappings without explicit access to the cipher. We develop ALICE (an Architecture for Learning Interpretable Cryptogram dEcipherment), a simple encoder-only Transformer that sets a new state-of-the-art for both accuracy and speed on this decryption problem. Surprisingly, ALICE generalizes to unseen ciphers after training on only ${\sim}1500$ unique ciphers, a minute fraction ($3.7 \times 10^{-24}$) of the possible cipher space. To enhance interpretability, we introduce a novel bijective decoding head that explicitly models permutations via the Gumbel-Sinkhorn method, enabling direct extraction of learned cipher mappings. Through early exit and probing experiments, we reveal how ALICE progressively refines its predictions in a way that appears to mirror common human strategies -- early layers place greater emphasis on letter frequencies, while later layers form word-level structures. Our architectural innovations and analysis methods are applicable beyond cryptograms and offer new insights into neural network generalization and interpretability.

cs.LG↗

arXiv2024

The Multimodal Universe: Enabling Large-Scale Machine Learning with 100TB of Astronomical Scientific Data

We present the MULTIMODAL UNIVERSE, a large-scale multimodal dataset of scientific astronomical data, compiled specifically to facilitate machine learning research. Overall, the MULTIMODAL UNIVERSE contains hundreds of millions of astronomical observations, constituting 100\,TB of multi-channel and hyper-spectral images, spectra, multivariate time series, as well as a wide variety of associated scientific measurements and "metadata". In addition, we include a range of benchmark tasks representative of standard practices for machine learning methods in astrophysics. This massive dataset will enable the development of large multi-modal models specifically targeted towards scientific applications. All codes used to compile the MULTIMODAL UNIVERSE and a description of how to access the data is available at https://github.com/MultimodalUniverse/MultimodalUniverse

astro-ph.IM↗

arXiv2024

The Well: a Large-Scale Collection of Diverse Physics Simulations for Machine Learning

Machine learning based surrogate models offer researchers powerful tools for accelerating simulation-based workflows. However, as standard datasets in this space often cover small classes of physical behavior, it can be difficult to evaluate the efficacy of new approaches. To address this gap, we introduce the Well: a large-scale collection of datasets containing numerical simulations of a wide variety of spatiotemporal physical systems. The Well draws from domain experts and numerical software developers to provide 15TB of data across 16 datasets covering diverse domains such as biological systems, fluid dynamics, acoustic scattering, as well as magneto-hydrodynamic simulations of extra-galactic fluids or supernova explosions. These datasets can be used individually or as part of a broader benchmark suite. To facilitate usage of the Well, we provide a unified PyTorch interface for training and evaluating models. We demonstrate the function of this library by introducing example baselines that highlight the new challenges posed by the complex dynamics of the Well. The code and data is available at https://github.com/PolymathicAI/the_well.

cs.LG↗

arXiv2024

Iron Snails: non-equilibrium dynamics and spiral abundance patterns

Galaxies are not in a dynamical steady state. They continually undergo perturbations, e.g., from infalling dwarf galaxies and dark-matter substructure. After a dynamical perturbation, stars phase mix towards a new steady state; in so doing they generally form spiral structures, such as spiral density waves in galaxy disks and the Gaia Snail observed in the vertical phase-space density in the solar neighborhood. Structures in phase-space density can be hard to measure accurately, because spatially varying selection effects imprint their own patterns on the density. However, stellar labels such as metallicity, or other element abundances, or stellar masses and ages, can be measured even in the face of complex or unknown spatial selection functions. We show that if the equilibrium galaxy has phase-space gradients in these labels, any perturbation that could raise a spiral wave in the phase-space density will raise a spiral wave in the distribution of labels as well. We work out the relationship between the spiral patterns in the density and in the labels. As an example, we analyze the Gaia Snail and show that its amplitude and dynamical age as derived from elemental abundances (mainly [Mg/Fe]) follow similar patterns to those derived from the phase-space density. Our best model dates the Snail's perturbation to about 400 Myr ago although we find significant variations with angular momentum in the best-fit age. Conceptually, the ideas presented here are related to Orbital Torus Imaging, chemical tagging, and other methods that use stellar labels to trace dynamics.

astro-ph.GA↗

arXiv2023

Multiscale Feature Attribution for Outliers

Machine learning techniques can automatically identify outliers in massive datasets, much faster and more reproducible than human inspection ever could. But finding such outliers immediately leads to the question: which features render this input anomalous? We propose a new feature attribution method, Inverse Multiscale Occlusion, that is specifically designed for outliers, for which we have little knowledge of the type of features we want to identify and expect that the model performance is questionable because anomalous test data likely exceed the limits of the training data. We demonstrate our method on outliers detected in galaxy spectra from the Dark Energy Survey Instrument and find its results to be much more interpretable than alternative attribution approaches.

cs.LG↗

arXiv2023

Outlier Detection in the DESI Bright Galaxy Survey

We present an unsupervised search for outliers in the Bright Galaxy Survey (BGS) dataset from the DESI Early Data Release. This analysis utilizes an autoencoder to compress galaxy spectra into a compact, redshift-invariant latent space, and a normalizing flow to identify low-probability objects. The most prominent outliers show distinctive spectral features such as irregular or double-peaked emission lines, or originate from galaxy mergers, blended sources, and rare quasar types, including one previously unknown Broad Absorption Line system. A significant portion of the BGS outliers are stars spectroscopically misclassified as galaxies. By building our own star model trained on spectra from the DESI Milky Way Survey, we have determined that the misclassification likely stems from the Principle Component Analysis of stars in the DESI pipeline. To aid follow-up studies, we make the full probability catalog of all BGS objects and our pre-trained models publicly available.

astro-ph.GA↗

arXiv2023

Disentangling Stellar Age Estimates from Galactic Chemodynamical Evolution

Stellar ages are key for determining the formation history of the Milky Way, but are difficult to measure precisely. Furthermore, methods that use chemical abundances to infer ages may entangle the intrinsic evolution of stars with the chemodynamical evolution of the Galaxy. In this paper, we present a framework for making probabilistic predictions of stellar ages, and then quantify the contribution of both stellar evolution and Galactic chemical evolution to those predictions using SHAP values. We apply this interpretable prediction framework to both a simulated Milky Way sample containing stars in a variety of evolutionary stages and an APOGEE-mocked sample of red clump stars. We find that in the former case, stellar evolution is the dominant driver for age estimates, while in the latter case, the more restricted evolutionary information causes the model to proxy ages through the chemical evolution model. We show that as a result of the use of non-intrinsic Galactic chemical information, trends estimated with the predicted ages, such as the age-metallicity relation, can deviate from the truth.

astro-ph.GA↗

arXiv2022

The Dragonfly Spectral Line Mapper: Design and First Light

The Dragonfly Spectral Line Mapper (DSLM) is the latest evolution of the Dragonfly Telephoto Array, which turns it into the world's most powerful wide-field spectral line imager. The DSLM will be the equivalent of a 1.6m aperture $f$/0.26 refractor with a built-in Integral Field Spectrometer, covering a five square degree field of view. The new telescope is designed to carry out ultra-narrow bandpass imaging of the low surface brightness universe with exquisite control over systematic errors, including real-time calibration of atmospheric variations in airglow. The key to Dragonfly's transformation is the "Filter-Tilter", a mechanical assembly which holds ultra-narrow bandpass interference filters in front of each lens in the array and tilts them to smoothly shift their central wavelength. Here we describe our development process based on rapid prototyping, iterative design, and mass production. This process has resulted in numerous improvements to the design of the DSLM from the initial pathfinder instrument, including changes to narrower bandpass filters and the addition of a suite of calibration filters for continuum light subtraction and sky line monitoring. Improvements have also been made to the electronics and hardware of the array, which improve tilting accuracy, rigidity and light baffling. Here we present laboratory and on-sky measurements from the deployment of the first bank of lenses in May 2022, and a progress report on the completion of the full array in early 2023.

astro-ph.IM↗

arXiv2022

The pathfinder Dragonfly Spectral Line Mapper: Pushing the limits for ultra-low surface brightness spectroscopy

The pathfinder Dragonfly Spectral Line Mapper is a distributed aperture telescope based off of the Dragonfly Telephoto Array with additional instrumentation (the Dragonfly "Filter-Tilter") to enable ultranarrow bandpass imaging. The pathfinder is composed of three redundant optical tube assemblies (OTAs) which are mounted together to form a single field of view imaging telescope (where the effective aperture diameter increases as the square-root of the number of OTAs). The pathfinder has been on sky from March 2020 to October 2021 equipped with narrowband filters to provide proof-of-concept imaging, surface brightness limit measurements, on sky testing, and observing software development. Here we describe the pathfinder telescope and the sensitivity limits reached along with observing methods. We outline the current limiting factors for reaching ultra-low surface brightnesses and present a comprehensive comparison of instrument sensitivities to low surface brightness line emission and other methods of observing the ultra-faint line emission from diffuse gas. Finally, we touch on plans for the upcoming 120-OTA Dragonfly Spectral Line Mapper, which is currently under construction.

astro-ph.IM↗

arXiv2021

The Mass of the Milky Way from the H3 Survey

The mass of the Milky Way is a critical quantity which, despite decades of research, remains uncertain within a factor of two. Until recently, most studies have used dynamical tracers in the inner regions of the halo, relying on extrapolations to estimate the mass of the Milky Way. In this paper, we extend the hierarchical Bayesian model applied in Eadie & Jurić (2019) to study the mass distribution of the Milky Way halo; the new model allows for the use of all available 6D phase-space measurements. We use kinematic data of halo stars out to $142~{\rm kpc}$, obtained from the H3 Survey and $\textit{Gaia}$ EDR3, to infer the mass of the Galaxy. Inference is carried out with the No-U-Turn sampler, a fast and scalable extension of Hamiltonian Monte Carlo. We report a median mass enclosed within $100~{\rm kpc}$ of $\rm M(<100 \; kpc) = 0.69_{-0.04}^{+0.05} \times 10^{12} \; M_\odot$ (68% Bayesian credible interval), or a virial mass of $\rm M_{200} = M(<216.2_{-7.5}^{+7.5} \; kpc) = 1.08_{-0.11}^{+0.12} \times 10^{12} \; M_\odot$, in good agreement with other recent estimates. We analyze our results using posterior predictive checks and find limitations in the model's ability to describe the data. In particular, we find sensitivity with respect to substructure in the halo, which limits the precision of our mass estimates to $\sim 15\%$.

astro-ph.GA↗

arXiv2021

Molecular gas in a gravitationally lensed galaxy group at $z = 2.9$

Most molecular gas studies of $z > 2.5$ galaxies are of intrinsically bright objects, despite the galaxy population being primarily "normal" galaxies with less extreme star formation rates. Observations of normal galaxies at high redshift provide a more representative view of galaxy evolution and star formation, but such observations are challenging to obtain. In this work, we present ALMA $\rm ^{12}CO(J = 3 \rightarrow 2)$ observations of a sub-millimeter selected galaxy group at $z = 2.9$, resulting in spectroscopic confirmation of seven images from four member galaxies. These galaxies are strongly lensed by the MS 0451.6-0305 foreground cluster at $z = 0.55$, allowing us to probe the molecular gas content on levels of $\rm 10^9-10^{10} \; M_\odot$. Four detected galaxies have molecular gas masses of $\rm (0.2-13.1) \times 10^{10} \; M_\odot$, and the non-detected galaxies have inferred molecular gas masses of $\rm < 8.0 \times 10^{10} \; M_\odot$. We compare these new data to a compilation of 546 galaxies up to $z = 5.3$, and find that depletion times decrease with increasing redshift. We then compare the depletion times of galaxies in overdense environments to the field scaling relation from the literature, and find that the depletion time evolution is steeper for galaxies in overdense environments than for those in the field. More molecular gas measurements of normal galaxies in overdense environments at higher redshifts ($z > 2.5$) are needed to verify the environmental dependence of star formation and gas depletion.

astro-ph.GA↗