arXiv subjects

Liang Chen

Publications and source records attributed to Liang Chen.

At least 163 records · Page 9Linked to original sources

arXiv2025

Measuring Diversity in Synthetic Datasets

Large language models (LLMs) are widely adopted to generate synthetic datasets for various natural language processing (NLP) tasks, such as text classification and summarization. However, accurately measuring the diversity of these synthetic datasets-an aspect crucial for robust model performance-remains a significant challenge. In this paper, we introduce DCScore, a novel method for measuring synthetic dataset diversity from a classification perspective. Specifically, DCScore formulates diversity evaluation as a sample classification task, leveraging mutual relationships among samples. We further provide theoretical verification of the diversity-related axioms satisfied by DCScore, highlighting its role as a principled diversity evaluation method. Experimental results on synthetic datasets reveal that DCScore enjoys a stronger correlation with multiple diversity pseudo-truths of evaluated datasets, underscoring its effectiveness. Moreover, both empirical and theoretical evidence demonstrate that DCScore substantially reduces computational costs compared to existing methods. Code is available at: https://github.com/bluewhalelab/dcscore.

cs.CL↗

arXiv2025

Vulnerability-Aware Alignment: Mitigating Uneven Forgetting in Harmful Fine-Tuning

Harmful fine-tuning (HFT), performed directly on open-source LLMs or through Fine-tuning-as-a-Service, breaks safety alignment and poses significant threats. Existing methods aim to mitigate HFT risks by learning robust representation on alignment data or making harmful data unlearnable, but they treat each data sample equally, leaving data vulnerability patterns understudied. In this work, we reveal that certain subsets of alignment data are consistently more prone to forgetting during HFT across different fine-tuning tasks. Inspired by these findings, we propose Vulnerability-Aware Alignment (VAA), which estimates data vulnerability, partitions data into "vulnerable" and "invulnerable" groups, and encourages balanced learning using a group distributionally robust optimization (Group DRO) framework. Specifically, VAA learns an adversarial sampler that samples examples from the currently underperforming group and then applies group-dependent adversarial perturbations to the data during training, aiming to encourage a balanced learning process across groups. Experiments across four fine-tuning tasks demonstrate that VAA significantly reduces harmful scores while preserving downstream task performance, outperforming state-of-the-art baselines.

cs.LG↗

arXiv2025

ROD: RGB-Only Fast and Efficient Off-road Freespace Detection

Off-road freespace detection is more challenging than on-road scenarios because of the blurred boundaries of traversable areas. Previous state-of-the-art (SOTA) methods employ multi-modal fusion of RGB images and LiDAR data. However, due to the significant increase in inference time when calculating surface normal maps from LiDAR data, multi-modal methods are not suitable for real-time applications, particularly in real-world scenarios where higher FPS is required compared to slow navigation. This paper presents a novel RGB-only approach for off-road freespace detection, named ROD, eliminating the reliance on LiDAR data and its computational demands. Specifically, we utilize a pre-trained Vision Transformer (ViT) to extract rich features from RGB images. Additionally, we design a lightweight yet efficient decoder, which together improve both precision and inference speed. ROD establishes a new SOTA on ORFD and RELLIS-3D datasets, as well as an inference speed of 50 FPS, significantly outperforming prior models.

cs.CV↗

arXiv2025

A Game-Theoretic Foundation for Bitcoin's Price: A Security-Utility Equilibrium

This paper introduces a structural game-theoretic model to value decentralized digital assets like Bitcoin. Instead of relying on speculative beliefs, it frames the asset's price within a Rational-Expectations Security-Utility Nash Equilibrium (RESUNE). This equilibrium is a fixed point where the market-clearing price dictates the hash rate through a free-entry mining model, which in turn endogenously sets the network's security. The security, defined as one minus the probability of a 51% attack, is determined via a global games model of attacker coordination, providing a unique and continuous security function. We prove the existence of a RESUNE and offer conditions for its uniqueness and stability. The model predicts that the stabilizing direct effect of price on demand must outweigh the potentially destabilizing feedback from price to security. The framework generates testable predictions, such as a protocol halving causing a contraction in both hash rate and price. A structural Vector Autoregression (VAR) model is proposed to test this mechanism. The model decomposes Bitcoin's value into transactional utility, security, and speculative components and explains the observed unidirectional causality from price to hash rate.

cs.CR↗

arXiv2025

Sub-Parsec Acceleration and Collimation of NGC 4261's Twin Jets

We report the first robust evidence for a co-spatial sub-parsec acceleration and collimation zone (ACZ) in the twin jets of the nearby low-luminosity active galactic nucleus (LLAGN) NGC 4261. This result is derived from multifrequency Very Long Baseline Array imaging, combined with the frequency-dependent properties of the radio core (core shift and core size) and jet kinematics determined from the jet-to-counterjet brightness ratio. By applying multiple analysis methods and incorporating results from the literature, we identify a parabolic-to-conical structural transition in both the jet and counterjet, with the transition occurring at $(1.23\pm0.24)$ pc or $(8.1\pm1.6)\times10^3 R_{\rm s}$ (Schwarzschild radii) for the jet and $(0.97\pm0.29)$ pc or $(6.4\pm1.9)\times10^3 R_{\rm s}$ for the counterjet. We also derive the jet velocity field at distances of $\sim (10^3-2\times10^4) R_{\rm s}$. While local kinematic variations are present, the jet shows an overall acceleration to relativistic speeds from $\sim 10^3$ to $\sim8\times10^3 R_{\rm s}$, with a maximum Lorentz factor of $Γ_{\rm max} \approx 2.6$. Beyond this region, the jet gradually decelerates to sub-relativistic speeds. These findings support the existence of a sub-parsec-scale ($\lesssim 1.5$ pc) ACZ in NGC 4261, where the jet is accelerated via magnetic-to-kinetic energy conversion while being confined by external pressure. A brief comparison with M 87 suggests that the ACZ in NGC 4261 may represent a scaled-down analogue of that in M 87. These results point towards a potential diversity in jet ACZ properties, emphasizing the importance of extending such studies to a broader AGN population to elucidate the physical mechanisms at play.

astro-ph.GA↗

arXiv2025

A Stochastic Gradient Descent Method for Globally Minimizing Nearly Convex Functions

This paper proposes a stochastic gradient descent method with an adaptive Gaussian noise term for the global minimization of nearly convex functions, which are nonconvex and possess multiple strict local minimizers. The noise term, independent of the gradient, is determined by the difference between the current function value and a lower bound estimate of the optimal value. In both probability space and state space, we show that the proposed algorithm converges linearly to a neighborhood of the global optimal solution. The size of this neighborhood depends on the variance of the gradient and the deviation between the estimated lower bound and the optimal value. In particular, when full gradient information is available and a sharp lower bound of the objective function is provided, the algorithm achieves linear convergence to the global optimum. Furthermore, we introduce a double-loop scheme that alternately updates the lower bound estimate and the optimization sequence, enabling convergence to a neighborhood of the global optimum that depends solely on the gradient variance. Numerical experiments on several benchmark problems demonstrate the effectiveness of the proposed algorithm.

math.OC↗

arXiv2025

From Equipartition to Curvature: The Spectral Evolution of 4FGL Blazars

We investigate the evolution of spectral energy distribution (SED) and underlying electron energy distribution (EED) by modeling the nearly simultaneous broadband spectra of selected bright 4FGL blazars, in the context of a combined cooling and stochastic acceleration scenario. We find that one-zone leptonic model with log-parabolic (LP) EED can successfully fit the GeV-TeV emission of blazars. The synchrotron frequency $ν_s$ of blazars mainly evolves due to variation of electron peak energy $γ_{3p}$. The BL Lac objects (BL Lacs) show a negative trend in the $ν_s- ν_s L_s$ SED plane, known as blazar sequence, that does not seem to be an artifact of Doppler boosting, but driven by the equipartition constraints. A positive correlation is found between the derived magnetic field $B$ and electron density $n_e$, whereas $n_e$ and $γ_{3p}$ negatively relate, as expected in an equipartition scenario. The flat spectrum radio quasars (FSRQs) deviate significantly from such a scenario, indicating their jet parameters should be varying independently. The synchrotron peak frequency $ν_s$ and its spectral curvature $b_s$ negatively correlate for all blazars, confirming the stochastic particle acceleration in blazar jets. However, blazars do not show the signature of hard-sphere acceleration, indicating that magnetic turbulence in the jets might be soft and physical conditions might be near to steady state, consistent with equipartition. Furthermore, for BL Lacs, the SED curvature $b_s$ and the EED curvature $r$ and nearly meet the theoretical relationship $r=5b_s$, whereas the FSRQs show large deviation due to poor constrain on $b_s$ due to presence of thermal component.

astro-ph.HE↗

arXiv2025

One Last Attention for Your Vision-Language Model

Pretrained vision-language models (VLMs), such as CLIP, achieve remarkable zero-shot performance, yet their downstream potential hinges on effective fine-tuning. Most adaptation methods typically focus on refining representation from separate modalities (text or vision) but neglect the critical role of their fused representations in the decision-making process, \emph{\ie} rational matrix that drives the final prediction. To bridge the gap, we propose a simple yet effective \textbf{R}ational \textbf{Ada}ptaion ({RAda}) to explicitly exploit the final fused representation during fine-tuning. RAda employs a learned mask, obtained from a lightweight attention layer attached at the end of a VLM, to dynamically calibrate the contribution of each element in the rational matrix, enabling targeted adjustments to the final cross-modal interactions without incurring costly modifications to intermediate features. Experiments in different settings (i.e., updating, or freezing pretrained encoders in adaptation, and test-time training that can only access the unlabeled test data) show that RAda serves as a versatile fine-tuning technique, improving the baseline with minimal code and performing comparably against current arts in most settings. Code is available at \href{https://github.com/khufia/RAda/tree/main}{github.com/khufia/RAda}.

cs.CV↗

arXiv2025

Style-Adaptive Detection Transformer for Single-Source Domain Generalized Object Detection

Single-source domain generalization (SDG) in object detection aims to develop a detector using only source domain data that generalizes well to unseen target domains. Existing methods are primarily CNN-based and improve robustness through data augmentation combined with feature alignment. However, these methods are limited, as augmentation is only effective when the synthetic distribution approximates that of unseen domains, thus failing to ensure generalization across diverse scenarios. While DEtection TRansformer (DETR) has shown strong generalization in domain adaptation due to global context modeling, its potential for SDG remains underexplored. To this end, we propose Style-Adaptive DEtection TRansformer (SA-DETR), a DETR-based detector tailored for SDG. SA-DETR introduces an online domain style adapter that projects the style representation of unseen domains into the source domain via a dynamic memory bank. This bank self-organizes into diverse style prototypes and is continuously updated under a test-time adaptation framework, enabling effective style rectification. Additionally, we design an object-aware contrastive learning module to promote extraction of domain-invariant features. By applying gating masks that constrain contrastive learning in both spatial and semantic dimensions, this module facilitates instance-level cross-domain contrast and enhances generalization. Extensive experiments across five distinct weather scenarios demonstrate that SA-DETR consistently outperforms existing methods in both detection accuracy and domain generalization capability.

cs.CV↗

arXiv2025

A Survey of the Evolution of Language Model-Based Dialogue Systems: Data, Task and Models

Dialogue systems (DS), including the task-oriented dialogue system (TOD) and the open-domain dialogue system (ODD), have always been a fundamental task in natural language processing (NLP), allowing various applications in practice. Owing to sophisticated training and well-designed model architecture, language models (LM) are usually adopted as the necessary backbone to build the dialogue system. Consequently, every breakthrough in LM brings about a shift in learning paradigm and research attention within dialogue system, especially the appearance of pre-trained language models (PLMs) and large language models (LLMs). In this paper, we take a deep look at the history of the dialogue system, especially its special relationship with the advancements of language models. Thus, our survey offers a systematic perspective, categorizing different stages in a chronological order aligned with LM breakthroughs, providing a comprehensive review of state-of-the-art research outcomes. What's more, we turn our attention to emerging topics and engage in a discussion on open challenges, providing valuable insights into the future directions for LLM-based dialogue systems. In summary, this survey delves into the dynamic interplay between language models and dialogue systems, unraveling the evolutionary path of this essential relationship. Through this exploration, we pave the way for a deeper comprehension of the field, guiding future developments in LM-based dialogue systems.

cs.CL↗

arXiv2025

Two-Component gamma-ray Emission Spectrum and X-Ray Polarization of the Radio Galaxy Pictor A

Pictor A is a $γ$-ray emitting radio galaxy and has a bright hotspot called WHS, located $\sim$4 arcmin away from the nucleus. In this work, we present an analysis of its 16-year Fermi-LAT data and report the Imaging X-ray Polarimetry Explorer (IXPE) observations for this source. Our analysis of the Fermi-LAT observations reveals evidence of two components in the average $γ$-ray spectrum of Pictor A, exhibiting a statistically significant hardening from $Γ_{\rm γ,1}=3.25\pm0.15$ to $Γ_{\rm γ,2}=1.81\pm0.07$ at a break energy of $2.46\pm0.09$ GeV. Notably, variability of $γ$-rays is evident in Pictor A, predominantly driven by the component below the break energy, while the component above the break energy remains stable. Furthermore, our analysis reveals that a power-law function provides an adequate fit for the high-flux-state spectrum, while a broken power-law function remains necessary to accurately model the low-flux-state spectrum. We suggest that the low-energy component originates from the nucleus, while the high-energy component primarily stems from WHS. The broadband spectral energy distributions of both nucleus and WHS can be well represented by a simple leptonic model, with both $γ$-ray components attributed to the synchrotron-self-Compton (SSC) process. Analysis of IXPE data provides upper limits on the polarization degree of $Π_{\rm X}<$6.6% for the nucleus and $Π_{\rm X}<$56.4% for the WHS within the 2--8 keV band. For the nucleus, this result aligns with X-ray emission originating from the SSC process. However, the upper limit of $Π_{\rm X}<$56.4% for WHS is insufficient to conclusively determine the X-ray emission mechanism in this region.

astro-ph.HE↗

arXiv2025

Calculation of Photocarrier Generation from Optical Absorption for Time-domain Simulation of Optoelectronic Devices

Photocarrier generation rate in optoelectronic materials is often calculated using the Poynting vector in the frequency domain. However, this approach is not accurate in time-domain simulations of photoconductive devices because the instantaneous Poynting vector does not distinguish between power flux densities of optical and low-frequency electromagnetic fields. The latter is generated by photocurrents and is not supposed to contribute to the photocarrier generation since the corresponding photon energy is smaller than the bandgap energy of the optoelectronic material. This work proposes an optical absorption-based model to accurately calculate the generation rate in time-domain simulations. The proposed approach considers the material dispersion near the optical frequency corresponding to the bandgap energy of the optoelectronic material and calculates the instantaneous optical absorption from the polarization current density associated with this dispersion model. Numerical examples show that the proposed method is more accurate than the Poynting vector-based approach in calculating the instantaneous optical absorption. The method is further validated against experimental results via simulations of a photoconductive device, where the Poynting vector-based approach results in divergent carrier densities when the low-frequency fields are strong.

physics.optics↗

arXiv2025

What Matters in LLM-generated Data: Diversity and Its Effect on Model Fine-Tuning

With the remarkable generative capabilities of large language models (LLMs), using LLM-generated data to train downstream models has emerged as a promising approach to mitigate data scarcity in specific domains and reduce time-consuming annotations. However, recent studies have highlighted a critical issue: iterative training on self-generated data results in model collapse, where model performance degrades over time. Despite extensive research on the implications of LLM-generated data, these works often neglect the importance of data diversity, a key factor in data quality. In this work, we aim to understand the implications of the diversity of LLM-generated data on downstream model performance. Specifically, we explore how varying levels of diversity in LLM-generated data affect downstream model performance. Additionally, we investigate the performance of models trained on data that mixes different proportions of LLM-generated data, which we refer to as synthetic data. Our experimental results show that, with minimal distribution shift, moderately diverse LLM-generated data can enhance model performance in scenarios with insufficient labeled data, whereas highly diverse generated data has a negative impact. We hope our empirical findings will offer valuable guidance for future studies on LLMs as data generators.

cs.CL↗

arXiv2025

Kimi-VL Technical Report

We present Kimi-VL, an efficient open-source Mixture-of-Experts (MoE) vision-language model (VLM) that offers advanced multimodal reasoning, long-context understanding, and strong agent capabilities - all while activating only 2.8B parameters in its language decoder (Kimi-VL-A3B). Kimi-VL demonstrates strong performance across challenging domains: as a general-purpose VLM, Kimi-VL excels in multi-turn agent tasks (e.g., OSWorld), matching flagship models. Furthermore, it exhibits remarkable capabilities across diverse challenging vision language tasks, including college-level image and video comprehension, OCR, mathematical reasoning, and multi-image understanding. In comparative evaluations, it effectively competes with cutting-edge efficient VLMs such as GPT-4o-mini, Qwen2.5-VL-7B, and Gemma-3-12B-IT, while surpassing GPT-4o in several key domains. Kimi-VL also advances in processing long contexts and perceiving clearly. With a 128K extended context window, Kimi-VL can process diverse long inputs, achieving impressive scores of 64.5 on LongVideoBench and 35.1 on MMLongBench-Doc. Its native-resolution vision encoder, MoonViT, further allows it to see and understand ultra-high-resolution visual inputs, achieving 83.2 on InfoVQA and 34.5 on ScreenSpot-Pro, while maintaining lower computational cost for common tasks. Building upon Kimi-VL, we introduce an advanced long-thinking variant: Kimi-VL-Thinking-2506. Developed through long chain-of-thought (CoT) supervised fine-tuning (SFT) and reinforcement learning (RL), the latest model exhibits strong long-horizon reasoning capabilities (64.0 on MMMU, 46.3 on MMMU-Pro, 56.9 on MathVision, 80.1 on MathVista, 65.2 on VideoMMMU) while obtaining robust general abilities. Code and models are publicly accessible at https://github.com/MoonshotAI/Kimi-VL.

cs.CV↗

arXiv2025

SaGIF: Improving Individual Fairness in Graph Neural Networks via Similarity Encoding

Individual fairness (IF) in graph neural networks (GNNs), which emphasizes the need for similar individuals should receive similar outcomes from GNNs, has been a critical issue. Despite its importance, research in this area has been largely unexplored in terms of (1) a clear understanding of what induces individual unfairness in GNNs and (2) a comprehensive consideration of identifying similar individuals. To bridge these gaps, we conduct a preliminary analysis to explore the underlying reason for individual unfairness and observe correlations between IF and similarity consistency, a concept introduced to evaluate the discrepancy in identifying similar individuals based on graph structure versus node features. Inspired by our observations, we introduce two metrics to assess individual similarity from two distinct perspectives: topology fusion and feature fusion. Building upon these metrics, we propose Similarity-aware GNNs for Individual Fairness, named SaGIF. The key insight behind SaGIF is the integration of individual similarities by independently learning similarity representations, leading to an improvement of IF in GNNs. Our experiments on several real-world datasets validate the effectiveness of our proposed metrics and SaGIF. Specifically, SaGIF consistently outperforms state-of-the-art IF methods while maintaining utility performance. Code is available at: https://github.com/ZzoomD/SaGIF.

cs.LG↗

arXiv2025

Reliable Transmission of LTP Using Reinforcement Learning-Based Adaptive FEC

Delay/Disruption Tolerant Networking (DTN) employs the Licklider Transmission Protocol (LTP) with Automatic Repeat reQuest (ARQ) for reliable data delivery in challenging interplanetary networks. While previous studies have integrated packet-level Forward Erasure Correction (FEC) into LTP to reduce retransmission time costs, existing static and delay-feedback-based dynamic coding methods struggle with highly variable and unpredictable deep space channel conditions. This paper proposes a reinforcement learning (RL)-based adaptive FEC algorithm to address these limitations. The algorithm utilizes historical feedback and system state to predict future channel conditions and proactively adjust the code rate. This approach aims to anticipate channel quality degradation, thereby preventing decoding failures and subsequent LTP retransmissions and improving coding efficiency by minimizing redundancy during favorable channel conditions. Performance evaluations conducted in simulated Earth-Moon and Earth-Mars link scenarios demonstrate this algorithm's effectiveness in optimizing data transmission for interplanetary networks. Compared to existing methods, this approach demonstrates significant improvement, with matrix decoding failures reduced by at least 2/3.

cs.NI↗

arXiv2025

A Dynamic Relaxation Framework for Global Solution of ACOPF

Solving the Alternating Current Optimal Power Flow (AC OPF) problem to global optimality remains challenging due to its nonconvex quadratic constraints. In this paper, we present a unified framework that combines static piecewise relaxations with dynamic cut-generation mechanism to systematically tighten the classic Second-Order Cone Programming (SOCP) relaxation to arbitrarily small conic violation, thus enabling the recovery of globally optimal solutions. Two static formulations, Pyramidal Relaxation (PR) and Quasi-Pyramidal Relaxation (QPR), are introduced to tighten each branch-flow second-order cone via a finite union of wedges, providing controllable accuracy. Their dynamic counterparts, Dynamic PR (DPR) and Dynamic QPR (DQPR), embed on-the-fly cut generation within a branch-and-cut solver to improve scalability. Convergence is further accelerated through warm starts and a lightweight local-search post-processing. Extensive experiments on benchmarks demonstrate effective elimination of conic violations and flexible trade-offs between solution accuracy and runtime. Practical guidelines are derived for selecting appropriate variants based on network size and accuracy requirements.

math.OC↗

arXiv2025

NTPP: Generative Speech Language Modeling for Dual-Channel Spoken Dialogue via Next-Token-Pair Prediction

Inspired by the impressive capabilities of GPT-4o, there is growing interest in enabling speech language models (SLMs) to engage in natural, fluid spoken interactions with humans. Recent advancements have led to the development of several SLMs that demonstrate promising results in this area. However, current approaches have yet to fully exploit dual-channel speech data, which inherently captures the structure and dynamics of human conversation. In this work, we systematically explore the use of dual-channel speech data in the context of modern large language models, and introduce a novel generative modeling paradigm, Next-Token-Pair Prediction (NTPP), to enable speaker-independent dual-channel spoken dialogue learning using decoder-only architectures for the first time. We evaluate our approach on standard benchmarks, and empirical results show that our proposed method, NTPP, significantly improves the conversational abilities of SLMs in terms of turn-taking prediction, response coherence, and naturalness. Moreover, compared to existing methods, NTPP achieves substantially lower inference latency, highlighting its practical efficiency for real-time applications.

cs.CL↗