arXiv subjects

Ke Zhang

Publications and source records attributed to Ke Zhang.

At least 127 records · Page 7Linked to original sources

arXiv2025

The ALMA Survey of Gas Evolution of PROtoplanetary Disks (AGE-PRO): III. Dust and Gas Disk Properties in the Lupus Star-forming Region

We present Band 6 and Band 7 observations of 10 Lupus disks around M3-K6 stars from the ALMA survey of Gas Evolution in PROtoplanetary disks (AGE-PRO) Large Program. In addition to continuum emission in both bands, our Band 6 setup covers the $\mathrm{{}^{12}CO}$, $\mathrm{{}^{13}CO}$ and $\mathrm{C^{18}O}\,J$=2-1 lines, while our Band 7 setup covers the $\mathrm{N_2H^+}\,J$=3-2 line. All of our sources are detected in $\mathrm{{}^{12}CO}$ and $\mathrm{{}^{13}CO}$, 7 out of 10 are detected in $\mathrm{C^{18}O}$, and 3 are detected in $\mathrm{N_2H^+}$. We find strong correlations between the CO isotopologue line fluxes and the continuum flux densities. With the exception of one disk, we also identify a strong correlation between the $\mathrm{C^{18}O}\,J$=2-1 and $\mathrm{N_2H^+}\,J$=3-2 fluxes, indicating similar CO abundances across this sample. For the two sources with well-resolved continuum and $\mathrm{{}^{12}CO}\,J$=2-1 images, we find that their gas-to-dust size ratio is consistent with the median value of $\sim 2$ inferred from a larger sample of Lupus disks. We derive dust disk masses from continuum flux densities. We estimate gas disk masses by comparing $\mathrm{C^{18}O}\,J$=2-1 line fluxes with those predicted by the limited grid of self-consistent disk models of Ruaud et al. (2022). A comparison of these mass estimates with those derived by Trapman et al. (2025), using a combination of CO isotopologue and $\mathrm{N_2H^+}$ line emission, shows that the masses are consistent with each other. Some discrepancies appear for small and faint disks, but they are still within the uncertainties. Both methods find gas disk masses increase with dust disk masses, and gas-to-dust mass ratios are between $10-100$ in the AGE-PRO Lupus sample.

astro-ph.EP↗

arXiv2025

Two-Stage Lithium Niobate Nonlinear Photonic Circuits for Low-Crosstalk and Broadband All Optical Wavelength Conversion

All optical wavelength converters (AOWCs) that can effectively and flexibly switch optical signals between different wavelength channels are essential elements in future optical fiber communications and quantum information systems. A promising strategy for achieving high-performance AOWCs is to leverage strong three-wave mixing processes in second-order nonlinear nanophotonic devices, specifically thin-film periodically poled lithium niobate (TF-PPLN) waveguides. By exploiting the advantages of sub-wavelength light confinement and dispersion engineering compared with their bulk counterparts, TF-PPLN waveguides provide a viable route for realizing highly efficient and broadband wavelength conversion. Nevertheless, most existing approaches rely on a single TF-PPLN device to perform both frequency doubling of the telecom pump and the wavelength conversion process, resulting in significant crosstalk between adjacent signal channels. Here, we address this challenge by demonstrating a two-stage TF-PPLN nonlinear photonic circuit that integrates a second-harmonic generation module, a signal wavelength conversion module, and multiple adiabatic directional coupler-based pump filters, on a single chip. By decoupling the two nonlinear processes and leveraging the high pump-filtering extinction ratio, we achieve low-crosstalk AOWC with a side-channel suppression ratio exceeding 25 dB, substantially surpassing the performance of single-stage devices. Furthermore, our device exhibits an ultra-broad conversion bandwidth of 110 nm and a relatively high conversion efficiency of -15.6 dB, making it an attractive solution for future photonic systems. The two-stage AOWC design shows promise for low-noise phase-sensitive amplification and quantum frequency conversion in future classical and quantum photonic systems.

physics.optics↗

arXiv2025

Details Matter for Indoor Open-vocabulary 3D Instance Segmentation

Unlike closed-vocabulary 3D instance segmentation that is often trained end-to-end, open-vocabulary 3D instance segmentation (OV-3DIS) often leverages vision-language models (VLMs) to generate 3D instance proposals and classify them. While various concepts have been proposed from existing research, we observe that these individual concepts are not mutually exclusive but complementary. In this paper, we propose a new state-of-the-art solution for OV-3DIS by carefully designing a recipe to combine the concepts together and refining them to address key challenges. Our solution follows the two-stage scheme: 3D proposal generation and instance classification. We employ robust 3D tracking-based proposal aggregation to generate 3D proposals and remove overlapped or partial proposals by iterative merging/removal. For the classification stage, we replace the standard CLIP model with Alpha-CLIP, which incorporates object masks as an alpha channel to reduce background noise and obtain object-centric representation. Additionally, we introduce the standardized maximum similarity (SMS) score to normalize text-to-proposal similarity, effectively filtering out false positives and boosting precision. Our framework achieves state-of-the-art performance on ScanNet200 and S3DIS across all AP and AR metrics, even surpassing an end-to-end closed-vocabulary method.

cs.CV↗

arXiv2025

Operationalizing AI for Good: Spotlight on Deployment and Integration of AI Models in Humanitarian Work

Publications in the AI for Good space have tended to focus on the research and model development that can support high-impact applications. However, very few AI for Good papers discuss the process of deploying and collaborating with the partner organization, and the resulting real-world impact. In this work, we share details about the close collaboration with a humanitarian-to-humanitarian (H2H) organization and how to not only deploy the AI model in a resource-constrained environment, but also how to maintain it for continuous performance updates, and share key takeaways for practitioners.

cs.CL↗

arXiv2025

SD-GS: Structured Deformable 3D Gaussians for Efficient Dynamic Scene Reconstruction

Current 4D Gaussian frameworks for dynamic scene reconstruction deliver impressive visual fidelity and rendering speed, however, the inherent trade-off between storage costs and the ability to characterize complex physical motions significantly limits the practical application of these methods. To tackle these problems, we propose SD-GS, a compact and efficient dynamic Gaussian splatting framework for complex dynamic scene reconstruction, featuring two key contributions. First, we introduce a deformable anchor grid, a hierarchical and memory-efficient scene representation where each anchor point derives multiple 3D Gaussians in its local spatiotemporal region and serves as the geometric backbone of the 3D scene. Second, to enhance modeling capability for complex motions, we present a deformation-aware densification strategy that adaptively grows anchors in under-reconstructed high-dynamic regions while reducing redundancy in static areas, achieving superior visual quality with fewer anchors. Experimental results demonstrate that, compared to state-of-the-art methods, SD-GS achieves an average of 60\% reduction in model size and an average of 100\% improvement in FPS, significantly enhancing computational efficiency while maintaining or even surpassing visual quality.

cs.GR↗

arXiv2025

Game-Theoretic Modeling of Vehicle Unprotected Left Turns Considering Drivers' Bounded Rationality

Modeling the decision-making behavior of vehicles presents unique challenges, particularly during unprotected left turns at intersections, where the uncertainty of human drivers is especially pronounced. In this context, connected autonomous vehicle (CAV) technology emerges as a promising avenue for effectively managing such interactions while ensuring safety and efficiency. Traditional approaches, often grounded in game theory assumptions of perfect rationality, may inadequately capture the complexities of real-world scenarios and drivers' decision-making errors. To fill this gap, we propose a novel decision-making model for vehicle unprotected left-turn scenarios, integrating game theory with considerations for drivers' bounded rationality. Our model, formulated as a two-player normal-form game solved by a quantal response equilibrium (QRE), offers a more nuanced depiction of driver decision-making processes compared to Nash equilibrium (NE) models. Leveraging an Expectation-Maximization (EM) algorithm coupled with a subtle neural network trained on precise microscopic vehicle trajectory data, we optimize model parameters to accurately reflect drivers' interaction-aware bounded rationality and driving styles. Through comprehensive simulation experiments, we demonstrate the efficacy of our proposed model in capturing the interaction-aware bounded rationality and decision tendencies between players. The proposed model proves to be more realistic and efficient than NE models in unprotected left-turn scenarios. Our findings contribute valuable insights into the vehicle decision-making behaviors with bounded rationality, thereby informing the development of more robust and realistic autonomous driving systems.

eess.SY↗

arXiv2025

The JDISC Survey: Linking the Physics and Chemistry of Inner and Outer Protoplanetary Disk Zones

Mid-infrared spectroscopy of protoplanetary disks provides a chemical inventory of gas within a few au, where planets are readily detected around older stars. With the JWST Disk Infrared Spectral Chemistry Survey (JDISCS), we explore demographic trends among 31 disks observed with MIRI (MRS) and with previous ALMA millimeter continuum imaging at high angular resolution (5-10 au). With these S/N $\sim$200-450 spectra, we report emission from H$_2$O, OH, CO, C$_2$H$_2$, HCN, CO$_2$, [Ne II], [Ne III], and [Ar II]. Emission from H$_2$O, OH and CO is nearly ubiquitous for low-mass stars, and detection rates of all molecules are higher than for similar disks observed with Spitzer-IRS. Slab model fits to the molecular emission lines demonstrate that emission from C$_2$H$_2$, HCN, and possibly CO$_2$ is optically thin; thus since column densities and emitting radii are degenerate, observations are actually sensitive to the total molecular mass. C$_2$H$_2$ and HCN emission also typically originate in a hotter region ($920^{+70}_{-130}$, $820^{+70}_{-130}$ K, respectively) than CO$_2$ ($600^{+200}_{-160}$ K). The HCN to cold H$_2$O luminosity ratios are generally smaller in smooth disks, consistent with more efficient water delivery via icy pebbles in the absence of large dust substructures. The molecular emission line luminosities are also correlated with mass accretion rates and infrared spectral indices, similar to trends reported from Spitzer-IRS surveys. This work demonstrates the power of combining multi-wavelength observations to explore inner disk chemistry as a function of outer disk and stellar properties, which will continue to grow as the sample of observed Class II systems expands in the coming JWST observation cycles.

astro-ph.SR↗

arXiv2025

Distributed satellite information networks: Architecture, enabling technologies, and trends

Driven by the vision of ubiquitous connectivity and wireless intelligence, the evolution of ultra-dense constellation-based satellite-integrated Internet is underway, now taking preliminary shape. Nevertheless, the entrenched institutional silos and limited, nonrenewable heterogeneous network resources leave current satellite systems struggling to accommodate the escalating demands of next-generation intelligent applications. In this context, the distributed satellite information networks (DSIN), exemplified by the cohesive clustered satellites system, have emerged as an innovative architecture, bridging information gaps across diverse satellite systems, such as communication, navigation, and remote sensing, and establishing a unified, open information network paradigm to support resilient space information services. This survey first provides a profound discussion about innovative network architectures of DSIN, encompassing distributed regenerative satellite network architecture, distributed satellite computing network architecture, and reconfigurable satellite formation flying, to enable flexible and scalable communication, computing and control. The DSIN faces challenges from network heterogeneity, unpredictable channel dynamics, sparse resources, and decentralized collaboration frameworks. To address these issues, a series of enabling technologies is identified, including channel modeling and estimation, cloud-native distributed MIMO cooperation, grant-free massive access, network routing, and the proper combination of all these diversity techniques. Furthermore, to heighten the overall resource efficiency, the cross-layer optimization techniques are further developed to meet upper-layer deterministic, adaptive and secure information services requirements. In addition, emerging research directions and new opportunities are highlighted on the way to achieving the DSIN vision.

cs.IT↗

arXiv2025

Optimal ${L^2}$ error estimates for 2D/3D incompressible Cahn--Hilliard--magnetohydrodynamic equations

This paper focuses on an optimal error analysis of a fully discrete finite element scheme for the Cahn--Hilliard--magnetohydrodynamic (CH-MHD) system. The method use the standard inf-sup stable Taylor--Hood/MINI elements to solve the Navier--Stokes equations, Lagrange elements to solve the phase field, and particularly, the Nédélec elements for solving the magnetic induction field. Suffering from the strong coupling and high nonlinearity, the previous works just provide suboptimal error estimates for phase field and velocity field in $L^{2}/Ł^2$-norm under the same order elements, and the suboptimal error estimates for magnetic induction field in $\H(\rm curl)$-norm. To this end, we utilize the Ritz, Stokes, and Maxwell quasi-projections to eliminate the low-order pollution of the phase field and magnetic induction field. In addition to the optimal $Ł^2$-norm error estimates, we present the optimal convergence rates for magnetic induction field in $\H(\rm curl)$-norm and for velocity field in $\H^1$-norm. Moreover, the unconditional energy stability and mass conservation of the proposed scheme are preserved. Numerical examples are illustrated to validate the theoretical analysis and show the performance of the proposed scheme.

math.NA↗

arXiv2025

Hints of planet formation signatures in a large-cavity disk studied in the AGE-PRO ALMA Large Program

Detecting planet signatures in protoplanetary disks is fundamental to understanding how and where planets form. In this work, we report dust and gas observational hints of planet formation in the disk around 2MASS-J16120668-301027, as part of the ALMA Large Program "AGE-PRO: ALMA survey of Gas Evolution in Protoplanetary disks". The disk was imaged with the Atacama Large Millimeter/submillimeter Array (ALMA) at Band 6 (1.3 mm) in dust continuum emission and four molecular lines: $^{12}$CO(J=2-1), $^{13}$CO(J=2-1), C$^{18}$O(J=2-1), and H$_2$CO(J=3$_{(3,0)}$-2$_{(2,0)}$). Resolved observations of the dust continuum emission (angular resolution of $\sim 150$ mas, 20 au) show a ring-like structure with a peak at $0.57 ^{\prime \prime}$ (75 au), a deep gap with a minimum at 0.24$^{\prime \prime}$ (31 au), an inner disk, a bridge connecting the inner disk and the outer ring, along with a spiral arm structure, and a tentative detection (to $3σ$) of a compact emission at the center of the disk gap, with an estimated dust mass of $\sim 2.7-12.9$ Lunar masses. We also detected a kinematic kink (not coincident with any dust substructure) through several $^{12}$CO channel maps (angular resolution $\sim$ 200 mas, 30 au), located at a radius of $\sim 0.875^{\prime \prime}$ (115.6 au). After modeling the $^{12}$CO velocity rotation around the protostar, we identified a tentative rotating-like structure at the kink location with a geometry similar to that of the disk. We discuss potential explanations for the dust and gas substructures observed in the disk, and their potential connection to signatures of planet formation.

astro-ph.EP↗

arXiv2025

The ALMA Survey of Gas Evolution of PROtoplanetary Disks (AGE-PRO): V. Protoplanetary gas disk masses

The evolution of the gas mass of planet-forming disks around young stars is crucial for our understanding of planet formation, yet it has proven hard to constrain observationally, due both to the difficulties of measuring gas masses and the lack of a homogeneous sample. Here we present a large grid of thermochemical models which we use to measure protoplanetary gas disk masses of AGE-PRO, the ALMA survey of Gas Evolution in PROtoplanetary disks. AGE-PRO covers a sample of 30 disks around similar spectral type (M3-K6) stars with ages between 0.1 and 10 Myr. Our approach is to simultaneously fit observations of CO isotopologues and N2H+, a complementary molecule produced when CO freezes out. We find that the median gas mass of the three regions decreases over time, from 7.0(+4.4,-2.6)x10^-3 Msun in Ophiuchus (<1 Myr) to 9.4(+5.4,-3.4)x10^-4 Msun for Lupus (~1-3 Myr) and 6.8(+5.1,-2.8)x10^-4 Msun for Upper Sco (~2-6 Myr), with ~1 dex scatter in gas mass in each region. We note that the gas mass distributions for Lupus and Upper Sco look very similar, which could be due to survivorship bias for the latter. The median bulk CO abundance in the CO emitting layer is found to be a factor ~10 lower than the ISM value but does not significantly change between Lupus and Upper Sco. From Lupus to Upper Sco the median gas-to-dust mass ratio increases by a factor ~3 from ~40 to ~120, suggesting efficient inward pebble drift and/or the formation of planetesimals.

astro-ph.EP↗

arXiv2025

The ALMA Survey of Gas Evolution of PROtoplanetary Disks (AGE-PRO): VII. Testing accretion mechanisms from disk population synthesis

The architecture of planetary systems depends on the evolution of the disks in which they form. In this work, we develop a population synthesis approach to interpret the AGE-PRO measurements of disk gas mass and size considering two scenarios: turbulence-driven evolution with photoevaporative winds and MHD disk-wind-driven evolution. A systematic method is proposed to constrain the distribution of disk parameters from the disk fractions, accretion rates, disk gas masses, and CO gas sizes. We find that turbulence-driven accretion with initially compact disks ($R_0 \simeq 5-20~$au), low mass-loss rates, and relatively long viscous timescales ($t_{ν,0} \simeq 0.4-3~$Myr or $α_{SS} \simeq 2-4 \times 10^{-4}$) can reproduce the disk fraction and gas sizes. However, the distribution of apparent disk lifetime defined as the $M_D/\dot{M}_*$ ratio is severely overestimated by turbulence-driven models. On the other hand, MHD wind-driven accretion can reproduce the bulk properties of the disk populations from Ophiuchus to Upper Sco assuming compact disks with an initial magnetization of about $β\simeq 10^5$ ($α_{DW} \simeq 0.5-1 \times 10^{-3}$) and a magnetic field that declines with time. More studies are needed to confirm the low masses found by AGE-PRO, notably for compact disks that question turbulence-driven accretion. The constrained synthetic disk populations can now be used for realistic planet population models to interpret the properties of planetary systems on a statistical basis.

astro-ph.EP↗

arXiv2025

The ALMA Survey of Gas Evolution of PROtoplanetary Disks (AGE-PRO): X. Dust Substructures, Disk Geometries, and Dust-disk Radii

We perform visibility fitting to the dust continuum Band 6 1.3 mm data of the 30 protoplanetary disks in the AGE-PRO ALMA Large Program. We obtain disk geometries, dust-disk radii, and azimuthally symmetric radial profiles of the intensity of the dust continuum emission. We examine the presence of continuum substructures in the AGE-PRO sample by using these radial profiles and their residuals. We detect substructures in 15 out of 30 disks. We report five disks with large ($>$15 au) inner dust cavities. The Ophiuchus Class I disks show dust-disk substructures in $\sim80\%$ of the resolved sources. This evidences the early formation of substructures in protoplanetary disks. A spiral is identified in IRS 63, hinting to gravitational instability in this massive disk. We compare our dust-disk brightness radial profiles with gas-disk brightness radial profiles and discuss colocal substructures in both tracers. In addition, we discuss the evolution of dust-disk radii and substructures across Ophiuchus, Lupus, and Upper Scorpius. We find that disks in Lupus and Upper Scorpius with large inner dust cavities have typical gas-disk masses, suggesting an abundance of dust cavities in these regions. The prevalence of pressure dust traps at later ages is supported by a potential trend with time with more disks with large inner dust cavities (or "transition disks") in Upper Scorpius and the absence of evolution of dust-disk sizes with time in the AGE-PRO sample. We propose this is caused by an evolutionary sequence with a high fraction of protoplanetary disks with inner protoplanets carving dust cavities.

astro-ph.EP↗

arXiv2025

The ALMA Survey of Gas Evolution of PROtoplanetary Disks (AGE-PRO): XI. Beam-corrected gas disk sizes from fitting 12CO moment zero maps

The inward drift of mm-cm sized pebbles in protoplanetary disks has become an important part of our current theories of planet formation and, more recently, planet composition as well. The gas-to-dust size ratio of protoplanetary disks can provide an important constraint on how pebbles have drifted inward provided that observational effects, especially resolution, can be accounted for. Here we present a method for fitting beam-convolved models to integrated intensity maps of line emission using the astropy python package and use it to fit 12 CO moment zero maps of ten Lupus and ten Upper Scorpius protoplanetary disks from the AGE-PRO ALMA Large Program, a sample of disks around M3-K6 stars that cover the ~1 to 6 Myr of gas disk evolution. From the unconvolved best fit models we measure the gas disk size (RCO,90%[model]), which we combine with the dust disk size (Rdust,90%[FRANK]) from continuum visibility fits from Vioque et al. (2025, in press.) to compute beam-corrected gas-to-dust size ratios. In our sample we find gas-to-dust size ratios between ~1 and ~5.5, with a median value of 2.78(+0.37,-0.32). Contrary to models of dust evolution that predict an increasing size ratio with time, we find that the younger disks in Lupus have similar (or even larger) median ratios (3.02(+0.33,-0.33)) than the older disks in Upper Sco (2.46(+0.53,-0.38)). A possible explanation to this discrepancy is that pebble drift is halted in dust traps combined with truncation of the gas disk by external photo-evaporation in Upper Sco, although survivorship bias could also play a role.

astro-ph.EP↗

arXiv2025

Weakly Supervised Temporal Action Localization via Dual-Prior Collaborative Learning Guided by Multimodal Large Language Models

Recent breakthroughs in Multimodal Large Language Models (MLLMs) have gained significant recognition within the deep learning community, where the fusion of the Video Foundation Models (VFMs) and Large Language Models(LLMs) has proven instrumental in constructing robust video understanding systems, effectively surmounting constraints associated with predefined visual tasks. These sophisticated MLLMs exhibit remarkable proficiency in comprehending videos, swiftly attaining unprecedented performance levels across diverse benchmarks. However, their operation demands substantial memory and computational resources, underscoring the continued importance of traditional models in video comprehension tasks. In this paper, we introduce a novel learning paradigm termed MLLM4WTAL. This paradigm harnesses the potential of MLLM to offer temporal action key semantics and complete semantic priors for conventional Weakly-supervised Temporal Action Localization (WTAL) methods. MLLM4WTAL facilitates the enhancement of WTAL by leveraging MLLM guidance. It achieves this by integrating two distinct modules: Key Semantic Matching (KSM) and Complete Semantic Reconstruction (CSR). These modules work in tandem to effectively address prevalent issues like incomplete and over-complete outcomes common in WTAL methods. Rigorous experiments are conducted to validate the efficacy of our proposed approach in augmenting the performance of various heterogeneous WTAL models.

cs.CV↗

arXiv2025

CLIP-AE: CLIP-assisted Cross-view Audio-Visual Enhancement for Unsupervised Temporal Action Localization

Temporal Action Localization (TAL) has garnered significant attention in information retrieval. Existing supervised or weakly supervised methods heavily rely on labeled temporal boundaries and action categories, which are labor-intensive and time-consuming. Consequently, unsupervised temporal action localization (UTAL) has gained popularity. However, current methods face two main challenges: 1) Classification pre-trained features overly focus on highly discriminative regions; 2) Solely relying on visual modality information makes it difficult to determine contextual boundaries. To address these issues, we propose a CLIP-assisted cross-view audiovisual enhanced UTAL method. Specifically, we introduce visual language pre-training (VLP) and classification pre-training-based collaborative enhancement to avoid excessive focus on highly discriminative regions; we also incorporate audio perception to provide richer contextual boundary information. Finally, we introduce a self-supervised cross-view learning paradigm to achieve multi-view perceptual enhancement without additional annotations. Extensive experiments on two public datasets demonstrate our model's superiority over several state-of-the-art competitors.

cs.CV↗

arXiv2025

ProCrop: Learning Aesthetic Image Cropping from Professional Compositions

Image cropping is crucial for enhancing the visual appeal and narrative impact of photographs, yet existing rule-based and data-driven approaches often lack diversity or require annotated training data. We introduce ProCrop, a retrieval-based method that leverages professional photography to guide cropping decisions. By fusing features from professional photographs with those of the query image, ProCrop learns from professional compositions, significantly boosting performance. Additionally, we present a large-scale dataset of 242K weakly-annotated images, generated by out-painting professional images and iteratively refining diverse crop proposals. This composition-aware dataset generation offers diverse high-quality crop proposals guided by aesthetic principles and becomes the largest publicly available dataset for image cropping. Extensive experiments show that ProCrop significantly outperforms existing methods in both supervised and weakly-supervised settings. Notably, when trained on the new dataset, our ProCrop surpasses previous weakly-supervised methods and even matches fully supervised approaches. Both the code and dataset will be made publicly available to advance research in image aesthetics and composition analysis.

cs.CV↗

arXiv2025

OSoRA: Output-Dimension and Singular-Value Initialized Low-Rank Adaptation

Fine-tuning Large Language Models (LLMs) has become increasingly challenging due to their massive scale and associated computational costs. Parameter-Efficient Fine-Tuning (PEFT) methodologies have been proposed as computational alternatives; however, their implementations still require significant resources. In this paper, we present OSoRA (Output-Dimension and Singular-Value Initialized Low-Rank Adaptation), a novel PEFT method for LLMs. OSoRA extends Low-Rank Adaptation (LoRA) by integrating Singular Value Decomposition (SVD) with learnable scaling vectors in a unified framework. It first performs an SVD of pre-trained weight matrices, then optimizes an output-dimension vector during training, while keeping the corresponding singular vector matrices frozen. OSoRA substantially reduces computational resource requirements by minimizing the number of trainable parameters during fine-tuning. Comprehensive evaluations across mathematical reasoning, common sense reasoning, and other benchmarks demonstrate that OSoRA achieves comparable or superior performance to state-of-the-art methods like LoRA and VeRA, while maintaining a linear parameter scaling even as the rank increases to higher dimensions. Our ablation studies further confirm that jointly training both the singular values and the output-dimension vector is critical for optimal performance.

cs.CL↗