arXiv subjects

Youjia Wu

Publications and source records attributed to Youjia Wu.

10 recordsLinked to original sources

arXiv2026

Helix: Evolutionary Reinforcement Learning for Open-Ended Scientific Problem Solving

Large language models (LLMs) with reasoning abilities have demonstrated growing promise for tackling complex scientific problems. Yet such tasks are inherently domain-specific, unbounded and open-ended, demanding exploration across vast and flexible solution spaces. Existing approaches, whether purely learning-based or reliant on carefully designed workflows, often suffer from limited exploration efficiency and poor generalization. To overcome these challenges, we present HELIX -- a Hierarchical Evolutionary reinforcement Learning framework with In-context eXperiences. HELIX introduces two key novelties: (i) a diverse yet high-quality pool of candidate solutions that broadens exploration through in-context learning, and (ii) reinforcement learning for iterative policy refinement that progressively elevates solution quality. This synergy enables the discovery of more advanced solutions. On the circle packing task, HELIX achieves state-of-the-art result with a sum of radii of 2.63598308 using only a 14B model. Across standard machine learning benchmarks, HELIX further surpasses GPT-4o with a carefully engineered pipeline, delivering an average F1 improvement of 5.95 points on the Adult and Bank Marketing datasets.

cs.LG↗

arXiv2026

Operator Learning with Domain Decomposition for Geometry Generalization in PDE Solving

Neural operators have become increasingly popular in solving \textit{partial differential equations} (PDEs) due to their superior capability to capture intricate mappings between function spaces over complex domains. However, the data-hungry nature of operator learning inevitably poses a bottleneck for their widespread applications. At the core of the challenge lies the absence of transferability of neural operators to new geometries. To tackle this issue, we propose operator learning with domain decomposition, a local-to-global framework to solve PDEs on arbitrary geometries. Under this framework, we devise an iterative scheme \textit{Schwarz Neural Inference} (SNI). This scheme allows for partitioning of the problem domain into smaller subdomains, on which local problems can be solved with neural operators, and stitching local solutions to construct a global solution. Additionally, we provide a theoretical analysis of the convergence rate and error bound. We conduct extensive experiments on several representative PDEs with diverse boundary conditions and achieve remarkable geometry generalization compared to alternative methods. These analysis and experiments demonstrate the proposed framework's potential in addressing challenges related to geometry generalization and data efficiency.

cs.LG↗

arXiv2025

From Cheap Geometry to Expensive Physics: Elevating Neural Operators via Latent Shape Pretraining

Industrial design evaluation often relies on high-fidelity simulations of governing partial differential equations (PDEs). While accurate, these simulations are computationally expensive, making dense exploration of design spaces impractical. Operator learning has emerged as a promising approach to accelerate PDE solution prediction; however, its effectiveness is often limited by the scarcity of labeled physics-based data. At the same time, large numbers of geometry-only candidate designs are readily available but remain largely untapped. We propose a two-stage framework to better exploit this abundant, physics-agnostic resource and improve supervised operator learning under limited labeled data. In Stage 1, we pretrain an autoencoder on a geometry reconstruction task to learn an expressive latent representation without PDE labels. In Stage 2, the neural operator is trained in a standard supervised manner to predict PDE solutions, using the pretrained latent embeddings as inputs instead of raw point clouds. Transformer-based architectures are adopted for both the autoencoder and the neural operator to handle point cloud data and integrate both stages seamlessly. Across four PDE datasets and three state-of-the-art transformer-based neural operators, our approach consistently improves prediction accuracy compared to models trained directly on raw point cloud inputs. These results demonstrate that representations from physics-agnostic pretraining provide a powerful foundation for data-efficient operator learning.

cs.LG↗

arXiv2025

SpecDis: Value added distance catalogue for 4 million stars from DESI Year-1 data

We present the SpecDis value added stellar distance catalog accompanying DESI DR1. SpecDis trains a feed-forward Neural Network (NN) with Gaia parallaxes and gets the distance estimates. To build up unbiased training sample, we do not apply selections on parallax error or signal-to-noise (S/N) of the stellar spectra, and instead we incorporate parallax error into the loss function. Moreover, we employ Principal Component Analysis (PCA) to reduce the noise and dimensionality of stellar spectra. Validated by independent external samples of member stars with precise distances from globular clusters (GCs), dwarf galaxies, stellar streams, combined with blue horizontal branch (BHB) stars, we demonstrate that our distance measurements show no significant bias up to 100kpc, and are much more precise than Gaia parallax beyond 7kpc. The median distance uncertainties are 23%, 19%, 11% and 7% for S/N $<$ 20, 20 $\leq$ S/N$<$ 60, 60 $\leq$ S/N $<$ 100 and S/N $\geq$ 100. Selecting stars with $\log g<3.8$ and distance uncertainties smaller than 25%, we have more than 74,000 giant candidates within 50kpc to the Galactic center and 1,500 candidates beyond this distance. Additionally, we develop a Gaussian mixture model to identify unresolvable equal-mass binaries by modeling the discrepancy between the NN-predicted and the geometric absolute magnitudes from Gaia parallaxes and identify 120,000 equal-mass binary candidates. Our final catalog provides distances and distance uncertainties for $>$ 4 million stars, offering a valuable resource for Galactic astronomy.

astro-ph.GA↗

arXiv2024

Reference Neural Operators: Learning the Smooth Dependence of Solutions of PDEs on Geometric Deformations

For partial differential equations on domains of arbitrary shapes, existing works of neural operators attempt to learn a mapping from geometries to solutions. It often requires a large dataset of geometry-solution pairs in order to obtain a sufficiently accurate neural operator. However, for many industrial applications, e.g., engineering design optimization, it can be prohibitive to satisfy the requirement since even a single simulation may take hours or days of computation. To address this issue, we propose reference neural operators (RNO), a novel way of implementing neural operators, i.e., to learn the smooth dependence of solutions on geometric deformations. Specifically, given a reference solution, RNO can predict solutions corresponding to arbitrary deformations of the referred geometry. This approach turns out to be much more data efficient. Through extensive experiments, we show that RNO can learn the dependence across various types and different numbers of geometry objects with relatively small datasets. RNO outperforms baseline models in accuracy by a large lead and achieves up to 80% error reduction.

cs.LG↗

arXiv2024

Preconditioning for Physics-Informed Neural Networks

Physics-informed neural networks (PINNs) have shown promise in solving various partial differential equations (PDEs). However, training pathologies have negatively affected the convergence and prediction accuracy of PINNs, which further limits their practical applications. In this paper, we propose to use condition number as a metric to diagnose and mitigate the pathologies in PINNs. Inspired by classical numerical analysis, where the condition number measures sensitivity and stability, we highlight its pivotal role in the training dynamics of PINNs. We prove theorems to reveal how condition number is related to both the error control and convergence of PINNs. Subsequently, we present an algorithm that leverages preconditioning to improve the condition number. Evaluations of 18 PDE problems showcase the superior performance of our method. Significantly, in 7 of these problems, our method reduces errors by an order of magnitude. These empirical findings verify the critical role of the condition number in PINNs' training.

cs.LG↗

arXiv2023

Novel constraints on fifth forces and ultralight dark sector with asteroidal data

We study for the first time the possibility of probing long-range fifth forces utilizing asteroid astrometric data, via the fifth force-induced orbital precession. We examine nine Near-Earth Object (NEO) asteroids whose orbital trajectories are accurately determined via optical and radar astrometry. Focusing on a Yukawa-type potential mediated by a new gauge field (dark photon) or a baryon-coupled scalar, we estimate the sensitivity reach for the fifth force coupling strength and mediator mass in the mass range $m \simeq (10^{-21}-10^{-15})\,{\rm eV}$, near the ``fuzzy'' dark matter region. Our estimated sensitivity is comparable to leading limits from equivalence principle tests, potentially exceeding these in a specific mass range. Furthermore, we set new constraints on Yukawa gravity. The fifth force-induced precession increases with the orbital semi-major axis in the limit of a small mass $m$, motivating the study of objects further away from the Sun. We discuss future prospects for extending our study to more than a million asteroids, including NEOs, main-belt asteroids, Hildas, and Jupiter Trojans, as well as trans-Neptunian objects and exoplanets.

hep-ph↗

arXiv2022

Dark Stars Powered by Self-Interacting Dark Matter

Dark matter annihilation might power the first luminous stars in the Universe. These types of stars, known as dark stars, could form in $(10^6\mathrm{-}10^8)\,M_\odot$ protohalos at redshifts $z \sim 20$, and they could be much more luminous and larger in size than ordinary stars powered by nuclear fusion. We investigate the formation of dark stars in the self-interacting dark matter (SIDM) scenario. We present a concrete particle physics model of SIDM that can simultaneously give rise to the observed dark matter density, satisfy constraints from astrophysical and terrestrial searches, and address the various small-scale problems of collisionless dark matter via the self-interactions. In this model, the power from dark matter annihilation is deposited in the baryonic gas in environments where dark stars could form. We further study the evolution of SIDM density profiles in the protohalos at $z \sim 20$. As the baryon cloud collapses due to the various cooling processes, the deepening gravitational potential can speed up gravothermal evolution of the SIDM halo, yielding sufficiently high dark matter densities for dark stars to form. We find that SIDM-powered dark stars can have similar properties, such as their luminosity and size, as dark stars predicted in collisionless dark matter models.

hep-ph↗

arXiv2021

Using Action Space Clustering to Constrain the Accretion History of Milky Way like Galaxies

In the currently favored cosmological paradigm galaxies form hierarchically through the accretion of numerous satellite galaxies. Since the satellites are much less massive than the host halo, they occupy a small fraction of the volume in action space defined by the potential of the host halo. Since actions are conserved when the potential of the host halo changes adiabatically, stars from an accreted satellite are expected to remain clustered in action space as the host halo evolves. In this paper, we identify accreted satellites in three Milky Way like disk galaxies from the cosmological baryonic FIRE-2 simulations by tracking satellite galaxies through simulation snapshots. We then try to recover these satellites by applying the cluster analysis algorithm Enlink to the orbital actions of accreted star particles in the present-day snapshot. We define several metrics to quantify the success of the clustering algorithm and use these metrics to identify well-recovered and poorly-recovered satellites. We plot these satellites in the infall time-progenitor mass (or stellar mass) space, and determine the boundaries between the well-recovered and poorly-recovered satellites in these two spaces with classification tree method. The groups found by Enlink are more likely to correspond to a real satellite if they have high significance, a quantity assigned by Enlink. Since cosmological simulations predict that most stellar halos have a population of insitu stars, we test the ability of Enlink to recover satellites when the sample is contaminated by 10-50% of insitu star particles, and show that most of the satellites well-recovered by Enlink in the absence of insitu stars, stay well-recovered even with 50% contamination. We thus expect that, in the future, cluster analysis in action space will be useful in upcoming data sets (e.g. Gaia) for identifying accreted satellites in the Milky Way.

astro-ph.GA↗

arXiv2019

Uncertainties in Direct Dark Matter Detection in Light of Gaia's Escape Velocity Measurements

Direct detection experiments have set increasingly stringent limits on the cross section for spin-independent dark matter-nucleon interactions. In obtaining such limits, experiments primarily assume the standard halo model (SHM) as the distribution of dark matter in our Milky Way. Three astrophysical parameters are required to define the SHM: the local dark matter escape velocity, the local dark matter density and the circular velocity of the sun around the center of the galaxy. This paper studies the effect of the uncertainties in these three astrophysical parameters on the XENON1T exclusion limits using the publicly available DDCalc code. We compare limits obtained using the widely assumed escape velocity from the RAVE survey and the newly calculated escape velocity by Monari $et$ $al.$ using Gaia data. Our study finds that the astrophysical uncertainties are dominated by the uncertainty in the escape velocity (independent of the best fit value) at dark matter masses below 6 GeV and can lead to a variation of nearly 6 orders of magnitude in the exclusion limits at 4 GeV. Above a WIMP mass of 6 GeV, the uncertainty becomes dominated by the local dark matter density, leading to uncertainties of factors of $\sim$10 (3) at 6 (15) GeV WIMP mass in the exclusion limits. Additionally, this work finds that the updated best fit value for the escape velocity based on Gaia data leads to only very minor changes to the effects of the astrophysical uncertainties on the XENON1T exclusion limits.

hep-ph↗