arXiv subjects

Lin Yan

Publications and source records attributed to Lin Yan.

At least 19 recordsLinked to original sources

arXiv2026

Follow-up of SN 2025wny I: Space-based Observations of the First Multiply-imaged Superluminous Supernova

We present space-based follow-up observations of the superluminous Type I supernova (SLSN-I) SN 2025wny at redshift $z_{SN}=2.0151$, gravitationally lensed by two galaxies at redshifts $z_{G1}=0.3755$ and $z_{G2}=0.3766$ into five resolved images. SN 2025wny is the first strongly lensed SLSN discovered and the first galaxy-scale lensed supernova for which both photometric and spectroscopic time-delay measurements are feasible. As such, it opens a new observational window for precision cosmology and the study of stellar explosions near the epoch of peak cosmic star formation. Our follow-up observations comprise two epochs of Hubble Space Telescope (HST) imaging, together with near-infrared imaging and spectroscopy obtained with the James Webb Space Telescope (JWST). From these data, we measure precise astrometry and multi-band photometry for the five resolved supernova images, the host galaxy, and the two deflecting galaxies. HST provides accurate relative image positions and rest-frame ultraviolet photometry, while JWST delivers complementary near-infrared imaging and spectroscopy probing the rest-frame optical at high signal-to-noise ratio. Together they yield a detailed characterization of both the lensing configuration and the supernova spectral energy distribution over a broad wavelength range. The data presented here provide the observational foundation for the accompanying analyses of the supernova properties, lens modeling, and time-delay cosmography, including the astrometric, photometric, and spectroscopic information required to measure $H_0$.

astro-ph.CO↗

arXiv2026

Follow-up of SN 2025wny II: Superluminous Supernova Physics at Cosmic Noon

SN 2025wny is a gravitationally lensed, hydrogen-poor superluminous supernova (SLSN-I) at z = 2.015. To date, it is the most extensively observed high-redshift core-collapse SN and has the most detailed rest-frame UV observations of any SLSN. We present densely sampled rest-frame UV-to-optical photometry and spectroscopy out to +80 d post-peak (rest frame) from several facilities, including JWST, Keck, VLT, Gemini, the Palomar 200-inch, the Fraunhofer Telescope at Wendelstein, and the Liverpool Telescope. Correcting for lensing magnification, SN 2025wny reaches a peak pseudo-bolometric luminosity of $L_{\rm peak}\gtrsim4\times10^{44}$ erg s$^{-1}$ over rest-frame 1500-4230 \r{A}, placing it within the luminosity range of typical SLSNe-I. SN 2025wny exhibits several unusual features, including a continuum excess and sharp spectral features in the FUV from +20-60 d that coincide with an FUV light-curve plateau and higher inferred blackbody temperatures. SN 2025wny's spectra also show little to no UV line blanketing, no obvious O II absorption despite high temperatures, and evidence for C II, H$\alpha$, and possible He I. Light-curve modeling suggests that SN 2025wny may require a hybrid or non-standard power source. This work provides some of the first detailed constraints on high-redshift SLSNe and establishes SN 2025wny as an essential spectral and photometric reference for identifying and interpreting high-redshift SLSNe discovered by Rubin and Roman.

astro-ph.CO↗

arXiv2026

Follow-up of SN 2025wny III: Spectroscopic Time-delay Measurements of a Strongly Gravitationally Lensed Superluminous Supernova

We present spatially resolved spectra and infer the time-delays between the multiple images of the strongly gravitationally lensed superluminous supernova (SLSN) 2025wny at z=2.015. SN 2025wny is the first known spatially resolved strongly lensed SLSN and provides a unique opportunity to measure lensing delays through the temporal evolution of supernova spectra. We present a spectroscopic dataset spanning several months, including spatially resolved spectra of images A, B, C, D, and E. We identify and measure the wavelength evolution of spectral features using Gaussian-process modeling. The time delays are inferred by jointly fitting the temporal evolution of the spectral features, yielding $\Delta t_{AB}=-10.3 \pm 2.3$, $\Delta t_{AC}=0.1 \pm 3.6$, $\Delta t_{AD}=-65.7 \pm 3.5$, and $\Delta t_{AE}=3.7 \pm 8.8$ days (68% confidence intervals). These are the among most precise time-delay measurements obtained for a lensed supernova to date, whether from spectroscopic or photometric methods. The longest delay ($\Delta t_{AD}$) is particularly well constrained, with a ~5% precision. Combined with the lens model presented by M\"ortsell et al. (2026), the spectroscopic time-delays give a Hubble constant $H_0 = 70.2^{+8.2}_{-6.1}$ km/s/Mpc. Our analysis demonstrates that spectroscopic evolution provides an independent and complementary route to time-delay measurements in lensed supernova systems, avoiding reliance on photometric light curves alone. As future surveys discover larger samples of lensed supernovae, spectroscopic time-delay measurements will provide an important avenue for precision cosmography.

astro-ph.CO↗

arXiv2026

Follow-up of SN 2025wny IV: Photometric Time-delay Measurements of a Strongly Lensed Superluminous Supernova

We present photometric time-delay measurements of SN 2025wny, the first strongly lensed Type I superluminous supernova (SLSN-I), discovered at $z = 2.015$. Time-delay measurements from strongly lensed supernovae provide an independent probe of cosmology and the Hubble constant, $H_0$, without reliance on the local distance ladder. Using multi-facility imaging data, we performed scene-modelling photometry to deblend four of the lensed images (A-D) and construct $grizJ$-band light curves. We modelled the resolved light curves with Gaussian process regression using GausSN (Hayes et al. 2024) to infer relative time delays and magnifications between the lensed images. We found that a constant magnification model provides a suboptimal description of the data, motivating a time-dependent sigmoid magnification model to account for evolving relative magnification of image A. We measured time delays of $\Delta t_{AB} = -10.6^{+2.2}_{-2.5}$ days and $\Delta t_{AC} = 1.2^{+2.7}_{-2.6}$ days (68% credible intervals), consistent with independent spectroscopic measurements from Johansson et al. (2026). Combining the photometric time delays with the lens model of M\"ortsell et al. (2026) gives $H_{0,\:\rm photo} = 80.5^{+26.4}_{-16.7}\;\rm km\,s^{-1}\,Mpc^{-1}$, while including the spectroscopic time delays as well yields $H_{0,\:\rm comb} = 70.8^{+8.2}_{-6.1}\;{\rm km\,s^{-1}\,Mpc^{-1}}$. Our results further demonstrate the potential of strongly lensed supernovae as independent probes of $H_0$.

astro-ph.CO↗

arXiv2026

Follow-up of SN 2025wny V: Lens Modelling and Cosmography of a Strongly Lensed Superluminous Supernova at $z = 2.015$ using Space Data

We present a lensing and cosmographic analysis of the strongly lensed Type I superluminous supernova SN 2025wny at redshift $z=2.015$, multiply imaged by two foreground galaxies at $z=0.376$. Using imaging obtained with the Hubble Space Telescope and the James Webb Space Telescope, we model the lens system with two elliptical power-law mass distributions and an external shear component. In addition to the supernova image positions, the modelling incorporates the surface-brightness distribution of the lensed host galaxy. The inferred Einstein radii are $\theta_{\rm E,1}\simeq 1.6$" and $\theta_{\rm E,2}\simeq 0.7$ - $0.8$", with broadly consistent results across all filters. After accounting for microlensing by stars in the lens galaxies, the posterior distribution spans total magnifications of approximately $\mu_{\rm tot}\sim 5$--$50$, with flux ratios of the multiple images consistent with observations. Combining the lens models with spectroscopically and photomerically measured time delays yields a filter-marginalized constraint of \[ H_0 = 66.7^{+7.6}_{-6.3}\; {\rm km\,s^{-1}\,Mpc^{-1}}, \] for a fiducial model with isothermal mass profiles. Allowing the density slopes of the lens galaxies to vary over a broad range results in \[ H_0 = 70.8^{+8.2}_{-6.1}\; {\rm km\,s^{-1}\,Mpc^{-1}}. \] These values are conditional on the adopted parameterization of the lens mass distribution, the assumed priors on the density slopes, and possible additional lensing contributions from the surrounding large-scale environment. We find that incorporating the currently available stellar kinematic measurements has only a modest effect on the inferred value of $H_0$. Future measurements of the lens-galaxy kinematics and a detailed characterization of the lens environment will further strengthen the utility of SN 2025wny as a cosmological probe.

astro-ph.CO↗

arXiv2026

Homology-Preserving Dimensionality Reduction via Adaptive Mapper and Landmark Isomap

As data becomes increasingly central across engineering and scientific disciplines, effective visualization is essential for interpreting complex, high-dimensional structures. Dimensionality reduction techniques project high-dimensional data into lower dimensions while aiming to preserve structural properties such as pairwise distances and local neighborhoods. In this paper, we focus on improving homological preservation, that is, the retention of topological features such as connected components and loops, which is critical for maintaining global shape and continuity. We first introduce AdaMapper, a Mapper-based algorithm that leverages persistence diagrams to guide both skeleton construction and landmark selection. AdaMapper incorporates an adaptive refinement strategy that automatically increases cover resolution in regions exhibiting topological loops. We then propose AdaHIsomap, which extends landmark Isomap by incorporating homology-informed landmark selection and augmenting it with random anchor points to better balance distance and homology preservation. We evaluate both methods on a diverse set of datasets, including high-dimensional point clouds, scientific simulations, networks, and image data, and benchmark their performance against state-of-the-art approaches.

cs.CG↗

arXiv2026

Identifying Gems from Roman RAPIDly

The Nancy Grace Roman Space Telescope (Roman), set for launch as early as September 2026, will conduct wide-field infrared imaging surveys with unprecedented spatial resolution and cadence, enabling the discovery of millions of astronomical transients. Hence, it is necessary to have automated pipelines for generating alerts in place so that the telescope can begin discovering reliable transients and variable objects soon after it is launched. However, no real Roman data currently exist, making the development of such pipelines difficult. In this work, we present a machine learning model $RuBR$ and a general methodology for distinguishing genuine transient and variable detections from spurious (bogus) detections within the RAPID pipeline. In particular, we present three models using this methodology: $RuBR_{comb}$ trained and tested on combined locally injected and OpenUniverse2024 transients, $RuBR_{loc}$ trained on locally injected transients and tested on OpenUniverse2024 transients, and $RuBR_{DA}$ that combines locally injected transients with a fraction of OpenUniverse2024 transients in domain-adaptation mode for training. This paves the way for strategies to adapt the $RuBR_{comb}$ model to real observations in the absence of any ground-truth labels during the early phases of the Roman mission. While the image differencing pipeline continues to be improved, our experimental results demonstrate the effectiveness of the proposed approach and its promise for robust real-bogus classification in the Roman era.

cs.LG↗

arXiv2026

Dual-Scale Temporal Fusion Reveals Structured Predictability in Subseasonal-to-Seasonal Temperature Prediction

Subseasonal-to-seasonal (S2S) temperature forecasts, spanning several weeks to a few months, are critically needed in agriculture practice, energy planning, and extreme-weather induced risk management, yet their reliability varies substantially across seasons and regions. Forecast skill is often attributed primarily to lead time, but this perspective does not fully explain the spatiotemporal patterns of predictability. Here we show that S2S predictability is organized across interacting temporal components, spatial heterogeneity, and large-scale pattern coherence, and that this structure can be explicitly characterized and exploited. We develop a dual-scale learning framework that separates calendar-aligned historical climate context from lead-time matched recent weather evolution, combining them through spatially adaptive fusion to enable stable temperature forecasts across the 30 to 90-day window. The learned fusion weights reveal that the balance between these two temporal scales shifts systematically with season and geography: during winter, interannual context dominates over high latitudes and complex terrain where forecast is the most difficult, while summer predictions reflect a more balanced temporal contribution across the domain. This spatially explicit reorganization of predictability, rather than simple lead-time decay, emerges as the primary determinant of forecast skill within the subseasonal window. Topology-aware structural constraints further improve spatial coherence of predicted temperature fields, stabilizing large-scale pattern organization particularly over complex terrain. These results reframe S2S predictability as a structured, multi-scale phenomenon, providing a more interpretable foundation for improving forecast systems and informing their use in practice.

cs.LG↗

arXiv2026

Halfway to the Peak: Kinematic Signatures of Stable Rotating Disks in Luminous Infrared Galaxies at z=0.5-0.6

We present a kinematic study of six infrared-luminous galaxies observed with the Mid-InfraRed Instrument Medium-Resolution Spectrometer (MIRI/MRS) onboard JWST. These galaxies lie at $z = 0.5$--$0.6$, midway between the present day and the peak of cosmic star formation. Our sample spans a range of star formation (SF) and active galactic nucleus (AGN) contributions to the mid-infrared emission. We characterize the dynamical state of these IR-luminous galaxies and assess how AGN activity influences the kinematics of the interstellar medium. Using mid-IR atomic lines, we map galaxy kinematics beyond the local Universe for the first time. The spatial resolution of MIRI/MRS (3.0 kpc for 0.46$\arcsec$ at z $\sim$ 0.55) allows us to resolve the internal kinematics of our targets. We compute kinematic maps in three different emission lines ([Ar II]6.99$\mu$m, [Ne II]12.81$\mu$m, and H$_2$ 0-0 S(5)6.91$\mu$m). Using the [Ar II]6.99$\mu$m kinematic maps, we derive rotation curves for these sources. All galaxies exhibit ordered rotation, with $V/\sigma \geq 2$, consistent with stable disks. Although some show minor disturbances, we find no strong evidence for recent major mergers or galaxy-wide ionized outflows. We find no correlation between $V/\sigma$ and AGN fraction, suggesting AGN activity does not significantly disrupt global kinematics or that disk disruption is not required to trigger AGN. However, galaxies with higher AGN fractions show elevated central dispersions, indicating localized turbulence, possibly due to AGN feedback, stellar feedback, accretion or bulge structure. These IR-luminous galaxies likely represent mature, rotationally supported disks, with AGN activation occurring after disk assembly.

astro-ph.GA↗

arXiv2026

A Natural $\gtrsim 100\times$ Telescope: Discovery of the Strongly Lensed Type II SN 2025mkn at $z=1.37$

We present the discovery of SN 2025mkn, a gravitationally lensed Type II supernova. First detected as a blue transient in ZTF, 0.83$^{\prime\prime}$ from a $z=0.42$ elliptical galaxy, follow-up SNIFS/UH2.2m and LRIS/Keck spectra revealed absorption lines at $z=1.371$. Later JWST NIRCam imaging shows that the bright transient is a close pair of point sources separated by $\sim 0.07^{\prime\prime}$, and a 30 times fainter counterimage opposite the lens, for which NIRSpec reveals strong H$\alpha$ emission also at $z=1.371$. The light curves and spectra are consistent with the Type II supernova source being magnified $\gtrsim 100$ times, with $\sim 250$ required to reconcile its luminosity with that of nearby events such as SN 2023ixf. Lens models are consistent with such high magnifications, and always show that the faint image arrived first (undetected in earlier ZTF imaging), consistent with the later spectral phase of this fainter image. A fourth image is also predicted and possibly detected in the NIRSpec data. Light-curve-based time-delay measurements are not possible due to the first image being the faintest; however, the resolved NIRSpec spectra offer a future opportunity for time-delay cosmography through supernova phase measurements.

astro-ph.CO↗

arXiv2026

Late-Time HST UV Detections Reveal Eruptive Mass Loss and Circumstellar Interaction in a Quarter of Stripped-Envelope Supernovae

We present HST WFC3/UVIS F275W near-UV imaging of 91stripped-envelope supernovae (SE SNe; Types IIb, Ib, Ic) from Snapshot program SNAP-16657, observed at phases of 270-1845 days (median 952 days) after first optical detection. We detect UV counterparts in 13 SE~SNe, of which 6 are classified as secure and 7 as ambiguous after comparison to nearby H\textsc{ii} regions, interpreting the secure sources as signatures of interaction with circumstellar material (CSM). Independent WISE W1/W2 light curves show $>300$ day mid-IR excesses in two of the secure UV sources, corroborating the interaction interpretation, and reveal two additional IR-only candidates without UV counterparts, indicating dust-obscured interaction episodes missed by the UV survey. A forward-modeling MCMC analysis using a physics-based CSM interaction model with three free parameters, the interaction fraction $f_\mathrm{CSM}$, shell mass $M_\mathrm{CSM}$, and thickness fraction $f_\mathrm{thick}$, yields $f_\mathrm{CSM} = 0.23^{+0.17}_{-0.09}$, $M_\mathrm{CSM} \approx 0.013~M_\odot$, and $f_\mathrm{thick} \approx 0.07$. The inferred thin-shell geometry implies an ejection duration of $\sim$6 yr for an outflow velocity of $300$ km s$^{-1}$, two to three orders of magnitude shorter than the thermal timescale of stable Roche-lobe overflow. This result disfavors steady binary mass transfer as the origin of the detected CSM and instead points to eruptive pre-supernova mass ejection in the final years before core collapse, either from wave-driven outbursts or from mass transfer triggered by late-stage progenitor re-expansion.

astro-ph.HE↗

arXiv2026

AT2024wpp: An Extremely Luminous Fast Ultraviolet Transient Powered by Accretion onto a Black Hole

We present the discovery of AT 2024wpp ("Whippet"), a fast and luminous 18cow-like transient. At a redshift of z=0.0868, revealed by Keck Cosmic Web Imager spectroscopy of its faint star-forming host, it is the fourth-nearest example of its class to date. Rapid identification of the source in the Zwicky Transient Facility data stream permitted ultraviolet-through-optical observations to be obtained prior to peak, allowing the first determination of the peak bolometric luminosity (2x10^45 erg/s), maximum photospheric radius (10^15 cm), and total radiated energy (10^51 erg) of an 18cow-like object. We present results from a comprehensive multiwavelength observing campaign, including a far-UV spectrum from the Cosmic Origins Spectrograph on the Hubble Space Telescope and deep imaging extending >100 days post-explosion from the Very Large Telescope, Hubble Space Telescope, Very Large Array, and Atacama Large Millimetre Array. We interpret the observations under a model in which a rapidly-accreting central engine blows a fast (~0.2c) wind into the surrounding medium and irradiates it with X-rays. The high Doppler velocities and intense ionization within this wind prevent identifiable spectroscopic features from appearing in the ejecta or in the surrounding circumstellar material. Weak H and He signatures do emerge in the spectra after 35 days in the form of double-peaked narrow lines. Each peak is individually narrow (full width ~3000 km/s) but the two components are separated by ~6600 km/s, indicating stable structures of denser material, possibly representing streams of tidal ejecta or an ablated companion star.

astro-ph.HE↗

arXiv2025

EP250827b/SN 2025wkm: An X-ray Flash-Supernova Powered by a Central Engine and Circumstellar Interaction

We present the discovery of EP250827b/SN 2025wkm, an X-ray Flash (XRF) discovered by the Einstein Probe (EP), accompanied by a broad-line Type Ic supernova (SN Ic-BL) at $z = 0.1194$. EP250827b possesses a prompt X-ray luminosity of $\sim 10^{45} \, \rm{erg \, s^{-1}}$, lasts over 1000 seconds, and has a peak energy $E_{\rm{p}} < 1.5$ keV at 90\% confidence. SN 2025wkm possesses a double-peaked optical light curve (LC), though its bolometric luminosity plateaus after its initial peak for $\sim 20$ days, consistent with a central engine injecting additional energy into the explosion. Its spectrum transitions from a blue to red continuum with clear blueshifted broad absorption features consistent with a SN Ic-BL classification. We do not detect any transient radio emission and rule out the existence of an on-axis, energetic jet $\gtrsim 10^{50}~$erg assuming a typical LGRB circumburst constant density ($n \approx 10^{-3}$--$10^{-1}~{\rm cm}^{-3}$) and microphysical parameters ($\epsilon_{\rm e} = 0.1$ and $\epsilon_{\rm B} = 0.01$). In the model we invoke, the collapse gives rise to a long-lived magnetar, potentially surrounded by an accretion disk. Magnetically--driven winds from the magnetar and the disk mix together and break out with a velocity $\sim 0.35c$ and interact with an extended circumstellar medium with radius $\sim 10^{13}$ cm, generating X-ray breakout emission through non-thermal free-free processes. The disk outflows and magnetar winds power blackbody photospheric emission as they cool adiabatically and thermalize, producing the first SN peak. The spin-down luminosity of the magnetar and radioactive decay of $^{56}$Ni powers the late-time emission. We end by discussing the landscape of XRF-SNe within the context of EP's recent discoveries.

astro-ph.HE↗

arXiv2025

Virtual Width Networks

We introduce Virtual Width Networks (VWN), a framework that delivers the benefits of wider representations without incurring the quadratic cost of increasing the hidden size. VWN decouples representational width from backbone width, expanding the embedding space while keeping backbone compute nearly constant. In our large-scale experiment, an 8-times expansion accelerates optimization by over 2 times for next-token and 3 times for next-2-token prediction. The advantage amplifies over training as both the loss gap grows and the convergence-speedup ratio increases, showing that VWN is not only token-efficient but also increasingly effective with scale. Moreover, we identify an approximately log-linear scaling relation between virtual width and loss reduction, offering an initial empirical basis and motivation for exploring virtual-width scaling as a new dimension of large-model efficiency.

cs.LG↗

arXiv2025

Discovery of SN 2025wny: a Strongly Gravitationally Lensed Superluminous Supernova at z = 2.01

We present the discovery of SN 2025wny (ZTF25abnjznp/GOTO25gtq) and spectroscopic classification of this event as the first gravitationally lensed Type I superluminous supernovae (SLSN-I). Deep ground-based follow-up observations resolves four images of the supernova with ~1.7" angular separation from the main lens galaxy, each coincident with the lensed images of a background galaxy seen in archival imaging of the field. Spectroscopy of the brightest point image shows narrow features matching absorption lines at a redshift of z = 2.011 and broad features matching those seen in superluminous SNe with Far-UV coverage. We infer a magnification factor of 20 to 50 for the brightest image in the system, based on photometric and spectroscopic comparisons to other SLSNe-I. SN 2025wny demonstrates that gravitationally-lensed SNe are in reach of ground-based facilities out to redshifts far higher than what has been previously assumed, and provide a unique window into studying distant supernovae, internal properties of dwarf galaxies, as well as for time-delay cosmography.

astro-ph.CO↗

arXiv2025

A JWST MIRI LRS Survey of 37 Massive Star-Forming Galaxies and AGN at Cosmic Noon -- Overview and First Results

We present a large spectroscopic survey with \textit{JWST}'s Mid-Infrared Instrument (MIRI) Low Resolution Spectrometer (LRS) targeting $37$ infrared-bright galaxies between $z=0.65-2.46$ with infrared luminosities $\log L_{\rm IR}/L_\odot>11.5$ and $\log M_*/M_\odot=10-11.5$. Targets were taken from a \textit{Spitzer} $24\,\mu$m-selected sample with archival spectroscopy from the Infrared Spectrograph (IRS) and include a mix of star-forming galaxies and dust-obscured AGN. By combining IRS with the increased sensitivity of LRS, we expand the range of spectral features observed between $5-30\,\mu$m for every galaxy in our sample. In this paper, we outline the sample selection, \textit{JWST} data reduction, 1D spectral extraction, and polycyclic aromatic hydrocarbon (PAH) feature measurements from $\lambda_{rest}=3.3-11.2\,\mu$m. In the \textit{JWST} spectra, we detect PAH emission features at $3.3-5.3\,\mu$m, as well as Paschen and Brackett lines. The $3.3\,\mu$m feature can be as bright as $1\%$ of the $8-1000\,\mu$m infrared luminosity and exhibits a tight correlation with the dust-obscured star-formation rate. We detect absorption features from CO gas, CO$_2$ ice, H$_2$O ice, and aliphatic dust. From the joint \textit{JWST} and \textit{Spitzer} analysis we find that the $11.3/3.3\,\mu$m PAH ratios are on-average three times higher than that of local luminous, infrared galaxies. This is interpreted as evidence that the PAH grains are larger at $z\sim1-2$. The size distribution may be affected by coagulation of grains due to high gas densities and low temperatures. These conditions are supported by the observation of strong water ice absorption at $3.05\,\mu$m, and can lower stellar radiative feedback as large PAHs transmit less energy per photon into the interstellar medium.

astro-ph.GA↗

arXiv2025

Risk-Sensitive RL for Alleviating Exploration Dilemmas in Large Language Models

Reinforcement Learning with Verifiable Rewards (RLVR) has proven effective for enhancing Large Language Models (LLMs) on complex reasoning tasks. However, existing methods suffer from an exploration dilemma: the sharply peaked initial policies of pre-trained LLMs confine standard RL algorithms to a narrow set of solutions, boosting single-solution accuracy (pass@1) but suppressing solution diversity and multi-solution performance (pass@k). As a result, RLVR often distills existing capabilities rather than discovering new reasoning strategies. To overcome this, we introduce a Risk-Sensitive Reinforcement Learning framework. Our approach employs a risk-seeking objective that interpolates between mean and maximum rewards, leading to a novel algorithm, Risk-Sensitive GRPO (RS-GRPO), which drives deeper exploration by amplifying learning from challenging prompts. Remarkably, RS-GRPO is simple to implement, requiring only minor code modifications. On six mathematical reasoning benchmarks and with five different LLMs, RS-GRPO consistently improves pass@k performance while maintaining or enhancing pass@1 accuracy.

cs.AI↗

arXiv2025

UI-TARS-2 Technical Report: Advancing GUI Agent with Multi-Turn Reinforcement Learning

The development of autonomous agents for graphical user interfaces (GUIs) presents major challenges in artificial intelligence. While recent advances in native agent models have shown promise by unifying perception, reasoning, action, and memory through end-to-end learning, open problems remain in data scalability, multi-turn reinforcement learning (RL), the limitations of GUI-only operation, and environment stability. In this technical report, we present UI-TARS-2, a native GUI-centered agent model that addresses these challenges through a systematic training methodology: a data flywheel for scalable data generation, a stabilized multi-turn RL framework, a hybrid GUI environment that integrates file systems and terminals, and a unified sandbox platform for large-scale rollouts. Empirical evaluation demonstrates that UI-TARS-2 achieves significant improvements over its predecessor UI-TARS-1.5. On GUI benchmarks, it reaches 88.2 on Online-Mind2Web, 47.5 on OSWorld, 50.6 on WindowsAgentArena, and 73.3 on AndroidWorld, outperforming strong baselines such as Claude and OpenAI agents. In game environments, it attains a mean normalized score of 59.8 across a 15-game suite-roughly 60% of human-level performance-and remains competitive with frontier proprietary models (e.g., OpenAI o3) on LMGame-Bench. Additionally, the model can generalize to long-horizon information-seeking tasks and software engineering benchmarks, highlighting its robustness across diverse agent tasks. Detailed analyses of training dynamics further provide insights into achieving stability and efficiency in large-scale agent RL. These results underscore UI-TARS-2's potential to advance the state of GUI agents and exhibit strong generalization to real-world interactive scenarios.

cs.AI↗