arXiv subjects

Boyuan Liu

Publications and source records attributed to Boyuan Liu.

At least 37 records · Page 2Linked to original sources

arXiv2026

Direct N-body simulations of rotating and extremely massive Population III star clusters

Aims. We present eight direct N-body simulations with NBODY6++GPU of extremely massive, initially rotating Population III star clusters with 1.01 x 10^5 stars. Methods. Our models include primordial binaries, a continuous initial mass function, differential rotation, tidal mass loss, updated fitting formulae for extremely massive metal-poor Population III stars, and general-relativistic merger recoil kicks. We assess their impact on cluster dynamics. Results. All runs form black holes below, within, and above the pair-instability gap, with multi-generation growth. Faster-rotating clusters core-collapse earlier; post-collapse clusters host a rotating, axisymmetric subsystem of intermediate-mass black holes (IMBHs) at the centre and an expanding halo of lower-mass objects. Pair-instability supernovae and compact-object formation at ~2-3 Myr sharply reduce total mass and a large fraction of the cluster's angular momentum. All Population III clusters in our simulations have the gravothermal-gravogyro catastrophe phase. Conclusions. We confirm two of the hypothesized formation channels of galactic nucleus seed black holes: gravitational runaway mergers of black holes and of Population III stars, which core-collapse into IMBHs thereafter. A higher initial star cluster bulk rotation correlates with earlier core collapse and, in the event counts reported here, with more coalescences and collisions, as well as lower retained (compact) binary abundances. Initial bulk rotation is a primary control parameter of cluster evolution: faster rotation accelerates early angular-momentum transport, gravothermal collapse, mass segregation, and amplifies post-collapse expansion, which also favours the formation of a compact central IMBH subsystem.

astro-ph.GA↗

arXiv2026

A black hole in a near-pristine galaxy 700 million years after the Big Bang

The recent discovery of a large number of massive black holes within the first two billion years after the Big Bang, as well as their peculiar properties, have been largely unexpected based on the extrapolation of the properties of luminous quasars. These findings have prompted the development of several theoretical models for the early formation and growth of black holes, which are, however, difficult to differentiate. We report the metallicity measurement around a gravitationally lensed massive black hole at redshift 7.04 (classified as a Little Red Dot), hosted in a galaxy with very low dynamical mass. The weakness of the [OIII]5007 emission line relative to the narrow H$β$ emission indicates extremely low metallicity, about $4\times 10^{-3}$ solar, and even more metal poor in the surrounding few 100 pc. We argue that such properties cannot be uncommon among accreting black holes around this early cosmic epoch. Explaining such a low chemical enrichment in a system that has developed a massive black hole is challenging for most theories. Models assuming heavy black hole seeds (such as Direct Collapse Black Holes) or super-Eddington accretion scenarios struggle to explain the observations, although they can potentially reproduce the observed properties in some cases. Models invoking "primordial black holes" (i.e. putative black holes formed shortly after the Big Bang) may potentially explain the low chemical enrichment associated with this black hole, although this class of models also requires further developments for proper testing.

astro-ph.GA↗

arXiv2026

The Cliff: A Metal-Poor Little Red Dot Hosting an Overmassive Black Hole at $z = 3.55$

JWST has revealed a large population of massive black holes (BHs) in the early Universe with unusual properties which mark them as distinct from low-redshift active galactic nuclei. Such findings have prompted the development of new models of BH formation and growth, and of their co-evolution with host galaxies. Linking the gas-phase metallicity of BH environments to seed masses is key to understanding which evolutionary pathways could explain the population of JWST-discovered BHs. We present new high-resolution JWST NIRSpec/IFU observations covering the rest-frame optical emission lines of a Little Red Dot (LRD) at $z=3.55$, known as The Cliff, from the `Red Unknowns: Bright Infrared Extragalactic Survey' (RUBIES). We find evidence for low metallicity ($Z=0.017\pm0.004 \ Z_\odot$) based on the low narrow-line [OIII]$\lambda5007$/H$β$ ratio, supported by the non-detection of low-ionisation emission lines such as [OII]$λ\lambda3727,3729$ and [NII]$λ\lambda6548,6583$. We find that the observed properties of The Cliff, including its overmassive BH, can be reproduced by some simulations of black hole growth and evolution down to $z\sim3.5$. However, these simulation runs require high seed masses ($10^4 - 10^5\ M_\odot$) and appear as rarely in the simulation volume as in the RUBIES survey volume over redshifts $3<z<4$, highlighting the unusual nature of The Cliff. Future simulations and numerical models will help to uncover how such a metal poor system managed to develop a massive black hole and persist to such low redshift.

astro-ph.GA↗

arXiv2026

Impact of initial mass function on the chemical evolution of high-redshift galaxies

Recent observations by the James Webb Space Telescope (JWST) have found evidence for an invariant relation between stellar mass, metallicity, and star formation rate up to $z\sim 8$ and its breakdown at higher redshifts. Understanding the underlying physics driving such correlations is thus crucial. Here, we explore the impact of the initial mass function (IMF) on the chemical evolution of high-redshift galaxies. Indeed, star formation and metal enrichment in galaxies are regulated by supernova (SN) explosions and metal yields from massive stars, which are sensitive to the high-mass end of the IMF. Using the semi-analytical galaxy evolution code \textsc{a-sloth}, we follow galactic baryon cycles along merger trees built from a high-resolution cosmological simulation. Stellar feedback is modeled with up-to-date stellar evolution tracks covering the full metallicity range ($Z \sim 10^{-11} - 0.03$) and a broad stellar mass range ($m_\star\sim2 - 600\ \rm M_\odot$), including metal yields from stellar winds, core-collapse SNe, (pulsational) pair-instability SNe, and Type Ia SNe. Assuming a Kroupa-like IMF with a varying upper mass limit $m_{\max}$, we find that only models with $m_{\max} \gtrsim 200\ \rm M_\odot$ can simultaneously reproduce the observed mass-metallicity-star formation rate relation and cosmic star formation history at $z\gtrsim 4$ owing to enhanced metal yields from pair-instability SNe. Our results confirm that very massive ($\gtrsim 200\ \rm M_\odot$) stars and pair-instability SNe play an important role in the star formation and chemical enrichment histories of high-$z$ galaxies. They also have profound implications for electromagnetic transients and gravitational-wave events.

astro-ph.GA↗

arXiv2026

Primordial Black Holes as Seeds for Extremely Overmassive AGN Observed by JWST

The James Webb Space Telescope (JWST) has recently identified Abell 2744-QSO1 as a compact, metal-poor, black hole (BH) dominated galaxy at $z\simeq 7$. This system exhibits an extreme black-hole-to-stellar mass ratio and unusually low metallicity, posing significant challenges to BH seeding models. Motivated by these discoveries, we perform high-resolution cosmological simulations with a massive primordial black hole (PBH; $M_{\rm BH}=5\times10^7\,M_\odot$) seed, incorporating for the first time a fully coupled treatment of PBH accretion, BH feedback, and Population~III/II star formation and stellar feedback. Although PBHs accelerate structure formation through the seed effect, the associated strong thermal feedback from the accretion delays the onset of star formation to $z\lesssim 10$, producing short, bursty episodes throughout the subsequent evolution. PBH-driven outflows expel enriched gas from the nucleus, while sustained inflows from the intergalactic medium continuously replenish pristine material. This feedback-regulated cycle naturally yields low accretion rates ($\dot{m}_{\rm BH}/\dot{m}_{\rm edd} \sim 1-10\%$), subsolar metallicities ($Z/Z_\odot\lesssim10^{-2}$) and extreme $M_{\rm BH}/M_\star$ ratios during both the initial star-forming phase and the subsequent quenching phases, in excellent agreement with JWST observations. Our results demonstrate that massive PBHs offer a viable pathway for forming the most extreme high-redshift systems, providing a physically motivated explanation for the extraordinary properties of Abell 2744-QSO1, as a sub-class of the broader population of JWST-discovered "little red dots".

astro-ph.GA↗

arXiv2026

GLIMPSE-DDT spectroscopic properties of faint-end galaxies at $z\sim6$: Towards first metal enrichment, dust production, and ionizing photon production

Ultra-faint galaxies at high-$z$ are fundamental elements of the early galaxy assembly, and spectroscopic characterization of this population is essential to understand the earliest galaxy evolution. Leveraging the ultra-deep JWST/NIRCam and NIRSpec observations of a gravitational lensing field of Abell S1063, taken as part of the GLIMPSE survey, we present spectroscopic properties of 16 galaxies fainter than $M_{\rm UV}=-17$ mag, including the metallicity, dust attenuation, and the ionizing photon production efficiency. The emission lines are generally quite strong, roughly half of which cannot be replicated with standard stellar populations and require an extreme ionizing source. We also identify relatively strong [OIII] emission lines from all sample galaxies, which indicates that the low-mass end of the mass-metallicity relation is extended down to $M_\star\sim10^6\ M_\odot$ at $z\sim6$. The strong [OIII] line detection from the lowest-mass galaxy among the sample ($M_\star\sim10^{5.6}\ M_\odot$) stands in contrast to recent reports of extremely metal-poor galaxy candidates at similar mass and redshift, suggesting that there could be two distinct pathways of the earliest metal enrichment as simulations have predicted. Interestingly, we detect both dust attenuation and galactic outflow in one of the sample galaxies with $M_\star=10^{6.6}\ M_\odot$ at $z=5.5$. All the dust, metal, and outflow contents in this galaxy can be consistently explained by supernovae (SNe), indicative of the key roles of SNe in the earliest galaxy assembly such as dust production, metal enrichment, stellar feedback, and baryon cycle.

astro-ph.GA↗

arXiv2026

Little Red Dots and their Progenitors from Direct Collapse Black Holes

The James Webb Space Telescope (JWST) has discovered a new population of objects, the Little Red Dots (LRDs), characterized by V-shaped spectra indicative of strong breaks around the Balmer limit and compact morphology that gave them their name. A popular explanation is that they are a sub-population of active galactic nuclei/supermassive black holes (AGN/SMBHs) predominantly found in the high-redshift Universe ($z\gtrsim3$). Similarly, direct collapse black holes (DCBHs), theorized to form from collapsing massive, extremely metal-poor gas clouds, have been invoked to explain high-redshift quasars, the most massive AGN sub-population. Here, we employ the semi-analytical code A-SLOTH to produce a population of DCBHs and compare them against observed LRD demographics and properties. Specifically, we compare the DCBH-seeded SMBH population against the standard stellar-remnant seeds and find that DCBH models agree better with observed LRD population statistics and host halo properties. Furthermore, for the most extreme and earliest LRD detections, interpreted to be systems with an AGN but little stellar component, DCBHs are able to reproduce the observed spectral shape and properties under multiple scenarios - high dust attenuation or AGN surrounded by dense gas - that have been proposed to explain the unique shape of LRD spectra. Even when super-Eddington accretion, invoked previously to explain the nature of LRDs, is enforced on stellar remnant seeds, the spectral characteristics of extreme LRDs cannot be reproduced. We emphasize the importance of gas-metallicity observations as an additional dimension besides the widely used SMBH-stellar mass ratios to further constrain the progenitors of LRDs.

astro-ph.GA↗

arXiv2025

Metal-poor single Wolf-Rayet stars: The interplay of optically thick winds and rotation

The Small Magellanic Cloud (SMC) hosts 12 known Wolf-Rayet (WR) stars, seven of which are apparently single. Their formation is a challenge for current stellar evolution models because line-driven winds are generally assumed to be quenched at a metallicity of Z < 0.004. Here, we present a set of mesa models of single stars with zero-age main sequence masses of 20 - 80 Msun considering different initial rotation speeds (Ω = 0 - 0.7 Ω_c), metallicities (Z = 0.002 - 0.0045), and wind mass-loss models (optically thin and thick winds). We show that if we account for optically thick winds, fast rotating (Ω = 0.6 Ω_c) single metal-poor O-type stars (with M > 20 Msun) shed their envelope and become WR stars even at the low metallicity of the SMC. The luminosity, effective temperature, evolutionary timescale, surface abundance, and rotational velocity of our simulated WR stars are compatible to the WRs observed in the SMC. We speculate that this scenario can also alleviate the excess of giant stars across the Humphreys-Davidson limit. Our results have key implications for black hole masses, (pair instability) supernova explosions, and other observable signatures.

astro-ph.SR↗

arXiv2025

Black Hole - Neutron Star and Binary Neutron Star Mergers from Population III and II stars

Population III (Pop.$~$III) stars are expected to be massive and to undergo minimal mass loss due to their lack of metals, making them ideal progenitors of black holes and neutron stars. Here, we investigate the formation and properties of binary neutron star (BNS) and black hole-neutron star (BHNS) mergers originating from Pop.$~$III stars, and compare them to their metal-enriched Population II (Pop.$~$II) counterparts, focusing on their merger rate densities (MRDs), primary masses and delay times. We find that, despite the high merger efficiency of Pop.$~$III BNSs and BHNSs, their low star formation rate results in a MRD at least one order of magnitude lower than that of Pop.$~$II stars. The MRD of Pop.$~$III BNSs peaks at redshift $z\sim15$, attaining a value $\mathcal{R}_{\rm BNS}(z\sim15) \sim 15\,\rm Gpc^{-3}\,yr^{-1}$, while the MRD of Pop.$~$III BHNSs is maximum at $z\sim13$, reaching a value $\mathcal{R}_{\rm BHNS}(z\sim13) \sim 2\,\rm Gpc^{-3}\,yr^{-1}$. Finally, we observe that the black hole masses of Pop.$~$III BHNS mergers have a nearly flat distribution with a peak at $\sim 20\,\rm M_{\odot}$ and extending up to $\sim 50\,\rm M_{\odot}$. Black holes in Pop.$~$II BHNS mergers show instead a peak at $\lesssim 15\,\rm M_{\odot}$. We consider these predictions in light of recent gravitational-wave observations in the local Universe, finding that a Pop.$~$III origin is preferred relative to Pop.$~$II for some events.

astro-ph.GA↗

arXiv2025

RIGEL: Feedback-regulated cloud-scale star formation efficiency in a simulated dwarf galaxy merger

Major mergers of galaxies are likely to trigger bursty star formation activities. The accumulation of dense gas and the boost of star formation efficiency (SFE) are considered to be the two main drivers of starbursts. However, it remains unclear how each process operates on the scale of individual star-forming clouds. Here, we present a high-resolution (2 Msun) RHD simulation of a gas-rich dwarf galaxy merger using the RIGEL model to investigate how mergers affect the properties of the structure of dense star-forming gas and the cloud-scale SFE. We tracked the evolution of sub-virial dense clouds in the simulation by mapping them across successive snapshots taken at intervals of 0.2 Myr. We find that the merger triggers a 130 fold increase in the SFR and shortens the galaxy-wide gas-depletion time by two orders of magnitude compared to those in two isolated galaxies. However, the depletion time of individual clouds and their lifetime distribution remained unchanged over the simulation period. The cloud life cycles and cloud-scale SFE are determined by the local stellar feedback rather than such environmental factors as tidal fields regardless of the merger process, and the integrated SFE ($ε_{\rm int}$) of clouds in complex environments remains well-described by an $ε_{\rm int}-Σ_{\rm tot}$ relation found in idealized isolated-cloud experiments. During the peak of the starburst, the media SFE was lower by only 0.17-0.33 dex compared to the value when the galaxies were not interacting. The merger boosts the SFR through the accumulation and compression of dense gas fueling star formation. Strong tidal torques assemble $>10^5$ Msun clouds, which seed massive star clusters. The average separation between star-forming clouds decreases during the merger, which in turn decreases the cloud--cluster spatial de-correlation from >1 kpc to 0.1 kpc depicted in tuning fork diagrams.

astro-ph.GA↗

arXiv2025

A Novel Formation Channel for Supermassive Black Hole Binaries in the Early Universe via Primordial Black Holes

We present a novel formation channel for supermassive black hole (SMBH) binaries in the early Universe, driven by primordial black holes (PBHs). Using high-resolution hydrodynamical simulations, we explore the role of massive PBHs ($m_{BH} \sim 10^6 M_\odot$) in catalyzing the formation of direct-collapse black holes (DCBHs), providing a natural in situ pathway for binary SMBH formation. PBHs enhance local overdensities, accelerate structure formation, and exert thermal feedback on the surrounding medium via accretion. Lyman-Werner (LW) radiation from accreting PBHs suppresses H$2$ cooling, shifting the dominant gas coolant to atomic hydrogen. When combined with significant baryon-dark matter streaming velocities ($v_{bχ} \gtrsim 0.8 σ_{bχ}$, where $σ_{bχ}$ is the root-mean-square streaming velocity), these effects facilitate the formation of dense, gravitationally unstable, atomically cooling gas clouds in the PBH's wake. These clouds exhibit sustained high inflow rates ($\dot{M}_{infall} \gtrsim 0.01 - 0.1 M_\odot yr^{-1}$), providing ideal conditions for DCBH formation from rapidly growing supermassive stars of $\sim 10^5 M_\odot$ at redshifts $z \sim 20 - 10$. The resulting systems form SMBH binaries with initial mass ratios $q \sim O(0.1)$ and separations of $\sim 10$ pc. Such PBH-DCBH binaries provide testable predictions for JWST and ALMA, potentially explaining select high-$z$ sources such as the Little Red Dots (LRDs), and represent gravitational-wave sources for future missions like LISA and TianQin-bridging early-Universe black hole physics, multi-messenger astronomy, and dark matter theory.

astro-ph.GA↗

arXiv2025

A direct black hole mass measurement in a Little Red Dot at the Epoch of Reionization

Recent discoveries of faint active galactic nuclei (AGN) at the redshift frontier have revealed a plethora of broad \Halpha emitters with optically red continua, named Little Red Dots (LRDs), which comprise 15-30\% of the high redshift broad line AGN population. Due to their peculiar spectral properties and X-ray weakness, modeling LRDs with standard AGN templates has proven challenging. In particular, the validity of single-epoch virial mass estimates in determining the black hole (BH) masses of LRDs has been called into question, with some models claiming that masses might be overestimated by up to 2 orders of magnitude, and other models claiming that LRDs may be entirely stellar in nature. We report the direct, dynamical BH mass measurement in a strongly lensed LRD at $z = 7.04$. The combination of lensing with deep spectroscopic data reveals a rotation curve that is inconsistent with a nuclear star cluster, yet can be well explained by Keplerian rotation around a point mass of 50 million Solar masses, consistent with virial BH mass estimates from the Balmer lines. The Keplerian rotation leaves little room for any stellar component in a host galaxy, as we conservatively infer $M_{\rm BH}/M_{*}>2$. Such a ''naked'' black hole, together with its near-pristine environment, indicates that this LRD is a massive black hole seed caught in its earliest accretion phase.

astro-ph.GA↗

arXiv2025

Omne-R1: Learning to Reason with Memory for Multi-hop Question Answering

This paper introduces Omne-R1, a novel approach designed to enhance multi-hop question answering capabilities on schema-free knowledge graphs by integrating advanced reasoning models. Our method employs a multi-stage training workflow, including two reinforcement learning phases and one supervised fine-tuning phase. We address the challenge of limited suitable knowledge graphs and QA data by constructing domain-independent knowledge graphs and auto-generating QA pairs. Experimental results show significant improvements in answering multi-hop questions, with notable performance gains on more complex 3+ hop questions. Our proposed training framework demonstrates strong generalization abilities across diverse knowledge domains.

cs.CL↗

arXiv2025

Effects of chemically homogeneous evolution of the first stars on the 21-cm signal and reionization

The first generation of stars, known as Population III (Pop III), played a crucial role in the early Universe through their unique formation environment and metal-free composition. These stars can undergo chemically homogeneous evolution (CHE) due to fast rotation, becoming more compact and hotter/bluer than their (commonly assumed) non-rotating counterparts. In this study, we investigate the impact of Pop III CHE on the 21-cm signal and cosmic reionization under various assumptions on Pop III star formation, such as their formation efficiency, initial mass function, and transition to metal-enriched star formation. We combine stellar spectra computed by detailed atmosphere models with semi-numerical simulations of Cosmic Dawn and the Epoch of Reionization ($z\sim 6-30$). The key effect of CHE arises from the boosted ionizing power of Pop III stars, which reduces the Pop III stellar mass density required to reproduce the observed Thomson scattering optical depth by a factor of $\sim 2$. Meanwhile, the maximum 21-cm global absorption signal is shallower by up to $\sim 15$ mK (11%), partly due to the reduced Lyman-band emission from CHE, and the large-scale ($k\sim 0.2\ \rm cMpc^{-1}$) power drops by a factor of a few at $z\gtrsim 25$. In general, the effects of CHE can be comparable to those of Pop III star formation parameters, showing an interesting interplay with distinct features in different epochs. These results highlight the importance of metal-free/poor stellar evolution in understanding the early Universe and suggest that future studies should consider joint constraints on the physics of star/galaxy formation and stellar evolution.

astro-ph.GA↗

arXiv2025

ContentV: Efficient Training of Video Generation Models with Limited Compute

Recent advances in video generation demand increasingly efficient training recipes to mitigate escalating computational costs. In this report, we present ContentV, an 8B-parameter text-to-video model that achieves state-of-the-art performance (85.14 on VBench) after training on 256 x 64GB Neural Processing Units (NPUs) for merely four weeks. ContentV generates diverse, high-quality videos across multiple resolutions and durations from text prompts, enabled by three key innovations: (1) A minimalist architecture that maximizes reuse of pre-trained image generation models for video generation; (2) A systematic multi-stage training strategy leveraging flow matching for enhanced efficiency; and (3) A cost-effective reinforcement learning with human feedback framework that improves generation quality without requiring additional human annotations. All the code and models are available at: https://contentv.github.io.

cs.CV↗

arXiv2025

The Emerging Black Hole Mass Function in the High-Redshift Universe

Observations with the James Webb Space Telescope (JWST) have identified an abundant population of supermassive black holes (SMBHs) already in place during the first few hundred million years of cosmic history. Most of them appear overmassive relative to the stellar mass in their host systems, challenging models of early black hole seeding and growth. Multiple pathways exist to explain their formation, including heavy seeds formed from direct collapse/supermassive stars or sustained super-Eddington accretion onto light stellar remnant seeds. We use the semi-analytical code A-SLOTH to predict the emerging SMBH mass function under physically motivated models for both light and heavy seed formation, to be compared with upcoming ultra-deep JWST surveys. We find that both pathways can reproduce observations at $z\sim5-6$, but have distinct features at higher redshifts of $z\sim10$. Specifically, JWST observations have the potential to constrain the fraction of efficiently accreting (super-Eddington) SMBHs, as well as the existence and prevalence of heavy seeds, in particular through ultra-deep observations of blank fields and/or gravitational lensing surveys. Such observations will provide key insights to understand the process of SMBH formation and evolution during the emergence of the first galaxies. We further emphasize the great promise of possible SMBH detections at $z\gtrsim 15$ with future JWST observations to break the degeneracy between light- and heavy-seed models.

astro-ph.GA↗

arXiv2025

How do Massive Primordial Black Holes Impact the Formation of the First Stars and Galaxies?

We investigate the impact of massive primordial black holes (PBHs; $m_{\rm BH}\sim 10^6~M_{\odot}$) on the star formation and first galaxy assembly process using high-resolution hydrodynamical simulations from $z = 1100$ to $z \sim 9$. We find that PBH accretion is self-regulated by feedback, suppressing mass growth unless feedback is weak. PBHs accelerate structure formation by seeding dark matter halos and gravitationally attracting gas, but strong feedback can delay cooling and suppress star formation. In addition, the presence of baryon-dark matter streaming creates an offset between the PBH location and the peaks induced in gas density, promoting earlier and more efficient star formation compared to standard $Λ$CDM. By $z \sim 10$, PBH-seeded galaxies form dense star clusters, with PBH-to-stellar mass ratios comparable to observed high-$z$ AGN like UHZ-1. Our results support PBHs as viable SMBH seeds but do not exclude alternative scenarios. We emphasize that PBH-seeding provides a natural explanation for some of the newly-discovered overmassive SMBHs at high redshift, in particular those with extreme ratios of BH-to-dynamical (virial) mass that challenge standard formation channels. Future studies with ultra-deep JWST surveys, the Roman Space Telescope, and radio surveys with facilities such as SKA and HERA will be critical in distinguishing PBH-driven SMBH growth from other pathways.

astro-ph.GA↗

arXiv2025

Towards Self-Improvement of Diffusion Models via Group Preference Optimization

Aligning text-to-image (T2I) diffusion models with Direct Preference Optimization (DPO) has shown notable improvements in generation quality. However, applying DPO to T2I faces two challenges: the sensitivity of DPO to preference pairs and the labor-intensive process of collecting and annotating high-quality data. In this work, we demonstrate that preference pairs with marginal differences can degrade DPO performance. Since DPO relies exclusively on relative ranking while disregarding the absolute difference of pairs, it may misclassify losing samples as wins, or vice versa. We empirically show that extending the DPO from pairwise to groupwise and incorporating reward standardization for reweighting leads to performance gains without explicit data selection. Furthermore, we propose Group Preference Optimization (GPO), an effective self-improvement method that enhances performance by leveraging the model's own capabilities without requiring external data. Extensive experiments demonstrate that GPO is effective across various diffusion models and tasks. Specifically, combining with widely used computer vision models, such as YOLO and OCR, the GPO improves the accurate counting and text rendering capabilities of the Stable Diffusion 3.5 Medium by 20 percentage points. Notably, as a plug-and-play method, no extra overhead is introduced during inference.

cs.CV↗