arXiv subjects

Xin Zhang

Publications and source records attributed to Xin Zhang.

At least 379 records · Page 21Linked to original sources

arXiv2025

Supervised Fine-Tuning or Contrastive Learning? Towards Better Multimodal LLM Reranking

In information retrieval, training reranking models mainly focuses on two types of objectives: metric learning (e.g. contrastive loss to increase the predicted scores on relevant query-document pairs) and classification (binary label prediction of relevance vs. irrelevance). For BERT-style encoders, various studies have shown that contrastive learning (CL) can be more effective than discriminative (classification) learning. However, for large language models (LLMs), classification via supervised fine-tuning (SFT), which predicts ''yes'' (resp. ''no'') token for relevant (resp. irrelevant) pairs, appears more promising as it aligns well with the generative nature of LLMs. This divergence raises a central question: which objective is intrinsically better suited to LLM-based reranking, and what mechanism underlies the difference? In this work, we conduct a comprehensive comparison and analysis between CL and SFT for reranking, taking the universal multimodal retrieval (UMR) as the experimental playground. We first decompose the objectives into two components: weight, which controls the magnitude of those updates, and direction, which guides the model updates, then present a unified framework for understanding their interactions. Through probing experiments, we find that SFT provides a substantially stronger weighting scheme than CL, whereas the preferred scoring direction shows no clear winner. Taken together, these results point to a consistent advantage of SFT over CL for LLM reranking. To further validate our findings, we conduct large-scale training with SFT and present new state-of-the-art rerankers on the MRB benchmark. We also provide ablations on SFT settings and expect our findings to benefit future research and applications in this area.

cs.CL↗

arXiv2025

Characterized behaviors of black hole thermodynamics in the supercritical region

The comprehension of universal thermodynamic behaviors in the supercritical region is crucial for examining the characteristics of black hole systems under high temperature and pressure. This study is devoted to the analysis of characteristic lines and crossover behaviors within the supercritical region. By making use of the free energy, we introduce three key thermodynamic quantities: scaled variance, skewness, and kurtosis. Our results demonstrate that the Widom line, associated with the maximal scaled variance, can effectively differentiate between small and large black hole-like subphases, each displaying distinct thermodynamic behaviors within the supercritical region. Furthermore, by utilizing quasinormal modes, we identify the Frenkel line, offering a dynamic perspective to distinguish between small and large black hole-like subphases. These contribute to a deeper comprehension of black hole subphases in the supercritical region, thus illuminating new facets of black hole thermodynamics.

gr-qc↗

arXiv2025

Prospects for probing dark matter particles and primordial black holes with the Square Kilometre Array using the 21 cm power spectrum at cosmic dawn

Probing the nature of dark matter (DM) remains an outstanding problem in modern cosmology. The 21 cm signal, as a sensitive tracer of neutral hydrogen during cosmic dawn, provides a unique means to investigate DM nature during this critical epoch. Annihilation and decay of DM particles, as well as Hawking radiation of primordial black holes (PBHs), can modify the thermal and ionization histories of the early universe, leaving distinctive imprints on the 21 cm power spectrum. Therefore, the redshifted 21 cm power spectrum serves as a powerful tool to investigate such DM processes. In this work, we systematically assess the potential of the upcoming Square Kilometre Array (SKA) to constrain DM and PBH parameters using the 21 cm power spectrum. Assuming $10,000$ hours of integration time, the SKA is projected to reach sensitivities of $\langleσv\rangle \leq 10^{-28}\,{\rm cm}^{3}\,{\rm s}^{-1}$ and $τ\geq 10^{28}\,{\rm seconds}$, for $10\,{\rm GeV}$ DM particles. It can also probe PBHs with masses of $10^{16}\,\mathrm{g}$ and abundances $f_{\mathrm{PBH}} \leq 10^{-6}$. These results indicate that the SKA could place constraints on DM annihilation, decay, and PBH Hawking radiation that are up to two to three orders of magnitude stronger than current limits. Furthermore, the SKA is expected to exceed existing bounds on sub-GeV DM and to probe Hawking radiation from PBHs with masses above $10^{17}\,{\rm g}$, which are otherwise inaccessible by conventional cosmological probes. Overall, the SKA holds great promise for advancing our understanding of both DM particles and PBHs, potentially offering new insights into the fundamental nature of DM.

astro-ph.CO↗

arXiv2025

Slitless Spectroscopy Source Detection Using YOLO Deep Neural Network

Slitless spectroscopy eliminates the need for slits, allowing light to pass directly through a prism or grism to generate a spectral dispersion image that encompasses all celestial objects within a specified area. This technique enables highly efficient spectral acquisition. However, when processing CSST slitless spectroscopy data, the unique design of its focal plane introduces a challenge: photometric and slitless spectroscopic images do not have a one-to-one correspondence. As a result, it becomes essential to first identify and count the sources in the slitless spectroscopic images before extracting spectra. To address this challenge, we employed the You Only Look Once (YOLO) object detection algorithm to develop a model for detecting targets in slitless spectroscopy images. This model was trained on 1,560 simulated CSST slitless spectroscopic images. These simulations were generated from the CSST Cycle 6 and Cycle 9 main survey data products, representing the Galactic and nearby galaxy regions and the high galactic latitude regions, respectively. On the validation set, the model achieved a precision of 88.6% and recall of 90.4% for spectral lines, and 87.0% and 80.8% for zeroth-order images. In testing, it maintained a detection rate >80% for targets brighter than 21 mag (medium-density regions) and 20 mag (low-density regions) in the Galactic and nearby galaxies regions, and >70% for targets brighter than 18 mag in high galactic latitude regions.

astro-ph.IM↗

arXiv2025

AirScape: An Aerial Generative World Model with Motion Controllability

How to enable agents to predict the outcomes of their own motion intentions in three-dimensional space has been a fundamental problem in embodied intelligence. To explore general spatial imagination capability, we present AirScape, the first world model designed for six-degree-of-freedom aerial agents. AirScape predicts future observation sequences based on current visual inputs and motion intentions. Specifically, we construct a dataset for aerial world model training and testing, which consists of 11k video-intention pairs. This dataset includes first-person-view videos capturing diverse drone actions across a wide range of scenarios, with over 1,000 hours spent annotating the corresponding motion intentions. Then we develop a two-phase schedule to train a foundation model--initially devoid of embodied spatial knowledge--into a world model that is controllable by motion intentions and adheres to physical spatio-temporal constraints. Experimental results demonstrate that AirScape significantly outperforms existing foundation models in 3D spatial imagination capabilities, especially with over a 50% improvement in metrics reflecting motion alignment. The project is available at: https://embodiedcity.github.io/AirScape/.

cs.RO↗

arXiv2025

Model-independent cosmological inference after the DESI DR2 data with improved inverse distance ladder

Recently, the baryon acoustic oscillations (BAO) measurements from the DESI survey have suggested hints of dynamical dark energy, challenging the standard $Λ$CDM model. In this work, we adopt an improved inverse distance ladder approach based on the latest cosmological data to provide a model-independent perspective, employing a global parametrization based on cosmic age (PAge). Our analysis incorporates DESI DR2 BAO measurements, cosmic chronometer (CC) data, and type Ia supernovae (SNe) observations from either the DESY5 or PantheonPlus datasets. For the DESY5+DESI DR2+CC datasets, we obtain $H_0 = 67.91 \pm 2.33~\mathrm{km~s^{-1}~Mpc^{-1}}$. This value is consistent with the Planck 2018 result, while shows $2.0 σ$ tension with the SH0ES measurement. Furthermore, by mapping specific cosmological models into PAge approximation parameter space $(p_{\mathrm{age}}, η)$, our model-independent analysis reveals a notable deviation from the $Λ\mathrm{CDM}$ model, as indicated by the DESY5 and DESI DR2 datasets. Finally, DESY5+DESI DR2+CC datasets provide nearly decisive evidence favoring the PAge model over the standard $Λ\mathrm{CDM}$ model. These findings highlight the need for further investigation into the expansion history to better understand the deviations from the $Λ\mathrm{CDM}$ model.

astro-ph.CO↗

arXiv2025

EpiCoder: Encompassing Diversity and Complexity in Code Generation

Existing methods for code generation use code snippets as seed data, restricting the complexity and diversity of the synthesized data. In this paper, we introduce a novel feature tree-based synthesis framework, which revolves around hierarchical code features derived from high-level abstractions of code. The feature tree is constructed from raw data and refined iteratively to increase the quantity and diversity of the extracted features, which captures and recognizes more complex patterns and relationships within the code. By adjusting the depth and breadth of the sampled subtrees, our framework provides precise control over the complexity of the generated code, enabling functionalities that range from function-level operations to multi-file scenarios. We fine-tuned widely-used base models to obtain EpiCoder series, achieving state-of-the-art performance on multiple benchmarks at both the function and file levels. In particular, empirical evidence indicates that our approach shows significant potential in the synthesizing of repository-level code data. Our code and data are publicly available at https://github.com/microsoft/EpiCoder.

cs.CL↗

arXiv2025

Teaching Your Models to Understand Code via Focal Preference Alignment

Preference learning extends the performance of Code LLMs beyond traditional supervised fine-tuning by leveraging relative quality comparisons. In existing approaches, a set of n candidate solutions is evaluated based on test case success rates, with the candidate demonstrating a higher pass rate being labeled as positive and its counterpart with a lower pass rate as negative. However, because this approach aligns entire failing code blocks rather than pinpointing specific errors, it lacks the granularity necessary to capture meaningful error-correction relationships. As a result, the model is unable to learn more informative error-correction patterns. To address these issues, we propose Target-DPO, a new preference alignment framework that mimics human iterative debugging to refine Code LLMs. Target-DPO explicitly locates error regions and aligns the corresponding tokens via a tailored DPO algorithm. To facilitate it, we introduce the CodeFlow dataset, where samples are iteratively refined until passing tests, with modifications capturing error corrections. Extensive experiments show that a diverse suite of Code LLMs equipped with Target-DPO achieves significant performance gains in code generation and improves on challenging tasks like BigCodeBench. In-depth analysis reveals that Target-DPO yields fewer errors. Code, model and datasets are in: https://github.com/JieWu02/Target-DPO.

cs.CL↗

arXiv2025

LogAction: Consistent Cross-system Anomaly Detection through Logs via Active Domain Adaptation

Log-based anomaly detection is a essential task for ensuring the reliability and performance of software systems. However, the performance of existing anomaly detection methods heavily relies on labeling, while labeling a large volume of logs is highly challenging. To address this issue, many approaches based on transfer learning and active learning have been proposed. Nevertheless, their effectiveness is hindered by issues such as the gap between source and target system data distributions and cold-start problems. In this paper, we propose LogAction, a novel log-based anomaly detection model based on active domain adaptation. LogAction integrates transfer learning and active learning techniques. On one hand, it uses labeled data from a mature system to train a base model, mitigating the cold-start issue in active learning. On the other hand, LogAction utilize free energy-based sampling and uncertainty-based sampling to select logs located at the distribution boundaries for manual labeling, thus addresses the data distribution gap in transfer learning with minimal human labeling efforts. Experimental results on six different combinations of datasets demonstrate that LogAction achieves an average 93.01% F1 score with only 2% of manual labels, outperforming some state-of-the-art methods by 26.28%. Website: https://logaction.github.io

cs.LG↗

arXiv2025

Generative Psycho-Lexical Approach for Constructing Value Systems in Large Language Models

Values are core drivers of individual and collective perception, cognition, and behavior. Value systems, such as Schwartz's Theory of Basic Human Values, delineate the hierarchy and interplay among these values, enabling cross-disciplinary investigations into decision-making and societal dynamics. Recently, the rise of Large Language Models (LLMs) has raised concerns regarding their elusive intrinsic values. Despite growing efforts in evaluating, understanding, and aligning LLM values, a psychologically grounded LLM value system remains underexplored. This study addresses the gap by introducing the Generative Psycho-Lexical Approach (GPLA), a scalable, adaptable, and theoretically informed method for constructing value systems. Leveraging GPLA, we propose a psychologically grounded five-factor value system tailored for LLMs. For systematic validation, we present three benchmarking tasks that integrate psychological principles with cutting-edge AI priorities. Our results reveal that the proposed value system meets standard psychological criteria, better captures LLM values, improves LLM safety prediction, and enhances LLM alignment, when compared to the canonical Schwartz's values.

cs.CL↗

arXiv2025

Traj-Transformer: Diffusion Models with Transformer for GPS Trajectory Generation

The widespread use of GPS devices has driven advances in spatiotemporal data mining, enabling machine learning models to simulate human decision making and generate realistic trajectories, addressing both data collection costs and privacy concerns. Recent studies have shown the promise of diffusion models for high-quality trajectory generation. However, most existing methods rely on convolution based architectures (e.g. UNet) to predict noise during the diffusion process, which often results in notable deviations and the loss of fine-grained street-level details due to limited model capacity. In this paper, we propose Trajectory Transformer, a novel model that employs a transformer backbone for both conditional information embedding and noise prediction. We explore two GPS coordinate embedding strategies, location embedding and longitude-latitude embedding, and analyze model performance at different scales. Experiments on two real-world datasets demonstrate that Trajectory Transformer significantly enhances generation quality and effectively alleviates the deviation issues observed in prior approaches.

cs.LG↗

arXiv2025

CottonSim: A vision-guided autonomous robotic system for cotton harvesting in Gazebo simulation

Cotton is a major cash crop in the United States, with the country being a leading global producer and exporter. Nearly all U.S. cotton is grown in the Cotton Belt, spanning 17 states in the southern region. Harvesting remains a critical yet challenging stage, impacted by the use of costly, environmentally harmful defoliants and heavy, expensive cotton pickers. These factors contribute to yield loss, reduced fiber quality, and soil compaction, which collectively threaten long-term sustainability. To address these issues, this study proposes a lightweight, small-scale, vision-guided autonomous robotic cotton picker as an alternative. An autonomous system, built on Clearpath's Husky platform and integrated with the CottonEye perception system, was developed and tested in the Gazebo simulation environment. A virtual cotton field was designed to facilitate autonomous navigation testing. The navigation system used Global Positioning System (GPS) and map-based guidance, assisted by an RGBdepth camera and a YOLOv8nseg instance segmentation model. The model achieved a mean Average Precision (mAP) of 85.2%, a recall of 88.9%, and a precision of 93.0%. The GPS-based approach reached a 100% completion rate (CR) within a $(5e-6)^{\circ}$ threshold, while the map-based method achieved a 96.7% CR within a 0.25 m threshold. The developed Robot Operating System (ROS) packages enable robust simulation of autonomous cotton picking, offering a scalable baseline for future agricultural robotics. CottonSim code and datasets are publicly available on GitHub: https://github.com/imtheva/CottonSim

cs.RO↗

arXiv2025

Phase transitions in a holographic superfluid model with non-linear terms beyond the probe limit

We study the holographic s-wave superfluid model with 4th and 6th power self-interaction terms $λ|ψ|^4$ and $τ|ψ|^6$ with considering the full back-reaction of the matter fields on the metric in the 3+1 dimensional bulk. The self-interaction terms are good at controlling the condensate to realize various phase transitions, such as the zeroth-order, first-order, and second-order phase transitions within the single condensate s-wave superfluid model. Therefore, in this work, we are able to investigate the influence of the back-reaction strength on the various phase transitions, including the zeroth and first order phase transitions. In addition, we confirm that the influence of the 4th and 6th power terms on the superfluid phase transition in the case of finite back-reaction are qualitative the same as in the probe limit, thus present universality. We also plot the special value $λ_s$ of the parameter $λ$ at different back-reaction strength, below which the condensate grows to an opposite direction and is important in controlling the order of the superfluid phase transitions. Comparing the influence of the back-reaction parameter and that of the higher-order nonlinear coefficients, we see that the back-reaction strength brings in both the effective couplings similar to the 4th power and 6th power terms.

hep-ph↗

arXiv2025

NAIPv2: Debiased Pairwise Learning for Efficient Paper Quality Estimation

The ability to estimate the quality of scientific papers is central to how both humans and AI systems will advance scientific knowledge in the future. However, existing LLM-based estimation methods suffer from high inference cost, whereas the faster direct score regression approach is limited by scale inconsistencies. We present NAIPv2, a debiased and efficient framework for paper quality estimation. NAIPv2 employs pairwise learning within domain-year groups to reduce inconsistencies in reviewer ratings and introduces the Review Tendency Signal (RTS) as a probabilistic integration of reviewer scores and confidences. To support training and evaluation, we further construct NAIDv2, a large-scale dataset of 24,276 ICLR submissions enriched with metadata and detailed structured content. Trained on pairwise comparisons but enabling efficient pointwise prediction at deployment, NAIPv2 achieves state-of-the-art performance (78.2% AUC, 0.432 Spearman), while maintaining scalable, linear-time efficiency at inference. Notably, on unseen NeurIPS submissions, it further demonstrates strong generalization, with predicted scores increasing consistently across decision categories from Rejected to Oral. These findings establish NAIPv2 as a debiased and scalable framework for automated paper quality estimation, marking a step toward future scientific intelligence systems. Code and dataset are released at sway.cloud.microsoft/Pr42npP80MfPhvj8.

cs.CL↗

arXiv2025

Dimension Reduction for Characterizing Sexual Dimorphism in Biomechanics of the Temporomandibular Joint

Sexual dimorphism is a critical factor in many biological and medical research fields. In biomechanics and bioengineering, understanding sex differences is crucial for studying musculoskeletal conditions such as temporomandibular disorder (TMD). This paper focuses on the association between the craniofacial skeletal morphology and temporomandibular joint (TMJ) related masticatory muscle attachments to discern sex differences. Data were collected from 10 male and 11 female cadaver heads to investigate sex-specific relationships between the skull and muscles. We propose a conditional cross-covariance reduction (CCR) model, designed to examine the dynamic association between two sets of random variables conditioned on a third binary variable (e.g., sex), highlighting the most distinctive sex-related relationships between skull and muscle attachments in the human cadaver data. Under the CCR model, we employ a sparse singular value decomposition algorithm and introduce a sequential permutation for selecting sparsity (SPSS) method to select important variables and to determine the optimal number of selected variables.

stat.ME↗

arXiv2025

Superradiant dark matter production from primordial black holes: Impact of multiple modes and gravitational wave emission

Rotating primordial black holes (PBHs) in the early universe can emit particles through superradiance, a process particularly efficient when the particle's Compton wavelength is comparable to the PBH's gravitational radius. Superradiance leads to an exponential growth of particle occupation numbers in gravitationally bound states. We present an analysis of heavy bosonic dark matter (DM) production through three gravitational mechanisms: Hawking radiation, superradiant instabilities, and ultraviolet (UV) freeze-in. We consider PBHs that evaporate before Big Bang Nucleosynthesis (BBN). For both scalar and vector DM, our analysis incorporates the evolution of a second superradiant mode. We demonstrate that the growth of a second superradiant mode causes the decay of the first mode, and thus the second mode cannot further enhance the DM abundance beyond that already achieved by the first mode. Our study also reveals that while superradiance generally enhances DM production, gravitational wave (GW) emission from the superradiant cloud may significantly modify this picture. For scalar DM, GW emission reduces the parameter space where superradiance effectively augments relic abundance. For vector DM, rapid GW emission from the superradiant cloud may yield relic abundances below those achieved through Hawking radiation alone. These findings demonstrate that multiple-mode effect and GW emission play critical roles in modeling DM production from PBHs in the early universe.

astro-ph.CO↗

arXiv2025

Fluid Antenna Meets RIS: Random Matrix Analysis and Two-Timescale Design for Multi-User Communications

The reconfigurability of fluid antenna systems (FASs) and reconfigurable intelligent surfaces (RISs) provides significant flexibility in optimizing channel conditions by jointly adjusting the positions of fluid antennas and the phase shifts of RISs. However, it is challenging to acquire the instantaneous channel state information (CSI) for both fluid antennas and RISs, while frequent adjustment of antenna positions and phase shifts will significantly increase the system complexity. To tackle this issue, this paper investigates the two-timescale design for FAS-RIS multi-user systems with linear precoding, where only the linear precoder design requires instantaneous CSI of the end-to-end channel, while the FAS and RIS optimization relies on statistical CSI. The main challenge comes from the complex structure of channel and inverse operations in linear precoding, such as regularized zero-forcing (RZF) and zero-forcing (ZF). Leveraging on random matrix theory (RMT), we first investigate the fundamental limits of FAS-RIS systems with RZF/ZF precoding by deriving the ergodic sum rate (ESR). This result is utilized to determine the minimum number of selected antennas to achieve a given ESR. Based on the evaluation result, we propose an algorithm to jointly optimize the antenna selection, regularization factor of RZF, and phase shifts at the RIS. Numerical results validate the accuracy of performance evaluation and demonstrate that the performance gain brought by joint FAS and RIS design is more pronounced with a larger number of users.

cs.IT↗

arXiv2025

Exploiting Block Coordinate Descent for Cost-Effective LLM Model Training

Training large language models typically demands extensive GPU memory and substantial financial investment, which poses a barrier for many small- to medium-sized teams. In this paper, we propose a full-parameter pre-training and fine-tuning framework based on block coordinate descent (BCD), enhanced with engineering optimizations, to enable efficient training of large-scale models on cost-effective RTX 4090, A100 and A800 GPU clusters. Under identical hardware configurations, we reduce the training cost of a 7B model to 33% on A100/A800 and only 2.6% on RTX 4090, compared to standard full-parameter training. It also enables large models previously restricted to A100 clusters to be trained on RTX 4090 without degrading performance. BCD achieves comparable or better accuracy than full-parameter and fine-tuning methods at most cases, with lower GPU consumption and improved hardware utilization.

cs.LG↗