Small Updates, Big Doubts: Does Parameter-Efficient Fine-tuning Enhance Hallucination Detection ?
研究探讨参数高效微调方法对大型语言模型在问答数据集上幻觉检测能力的影响，通过多种检测方法评估，发现该方法能显著增强幻觉检测性能。
1 citationsRead paper
Selected work
Representative Papers
Discovering Dual-Origin Slow Wind from Solar Orbiter with Self-Supervised Contrastive Learning
该研究使用自监督对比学习方法，通过Solar-CDC框架区分太阳慢速风的双起源问题。
0 citationsRead paper
Governing Agentic AI in FinTech
This study addresses the accountability challenges arising from deploying agent-based AI systems in finance when critical decision-making authority is granted without adequate governance, particularly due to insufficient verifiability. The authors propose a multi-layered governance framework that, for the first time, treats verifiability as a core constraint, introducing the concept of a “verifiability gap” and modeling reproducibility as an evidence-dependent governance profile. Through a three-phase empirical study comparing locally hosted large models with commercial frontier models, they systematically evaluate how temperature, top_p, top_k, and random seeds affect decision reproducibility, while also analyzing audit latency via execution logs and architectural configurations. Results show that under strict controls, local models achieve 320/320 exact reproductions and hosted models 959/960; however, architectural orchestration significantly influences outcomes, and historical decisions remain difficult to reconstruct—even with deterministic models—demonstrating that verifiability does not automatically improve with model scale.
0 citationsRead paper
Compactly supported radial basis functions as probability density functions
This work addresses the underexplored use of compactly supported radial basis functions (RBFs) in probabilistic density modeling by systematically formulating them as parametric probability density functions. Specifically, it introduces a mixture density model based on the Wendland 𝒞² kernel and derives closed-form expressions for key statistical quantities—including moments and cumulative distribution functions—under both truncated and untruncated support. An incremental learning algorithm is proposed, combining k-means for centroid initialization with stochastic gradient descent for parameter optimization. Experimental results demonstrate that the proposed method achieves likelihood performance and model complexity comparable to Gaussian mixture models on both synthetic and real-world datasets, while enabling exact computation of distributional properties owing to its compact support and analytical tractability.
0 citationsRead paper
An Evaluation of ABR Switching for Time-Shifted Clients in MoQ
This study addresses the vulnerability of MoQ’s default bitrate switching mechanism to playback interruptions under network congestion, which particularly degrades the time-shifted streaming experience. The authors evaluate SWITCH-style adaptive bitrate (ABR) algorithms in a Mininet emulation environment, demonstrating for the first time that standard ABR algorithms—without modification—can effectively support MoQ time-shifted playback. Experimental results reveal a counterintuitive yet significant increase in throughput following rebuffering events, suggesting that existing ABR strategies can directly enhance quality of service for time-shifted scenarios. These findings not only validate the immediate applicability of conventional ABR logic to MoQ-based time-shifted streaming but also highlight key directions for optimizing ABR switching behavior within the MoQ framework.
0 citationsRead paper
Recent publications
Latest Papers
Discovering Dual-Origin Slow Wind from Solar Orbiter with Self-Supervised Contrastive Learning
该研究使用自监督对比学习方法，通过Solar-CDC框架区分太阳慢速风的双起源问题。
0 citationsRead paper
Governing Agentic AI in FinTech
This study addresses the accountability challenges arising from deploying agent-based AI systems in finance when critical decision-making authority is granted without adequate governance, particularly due to insufficient verifiability. The authors propose a multi-layered governance framework that, for the first time, treats verifiability as a core constraint, introducing the concept of a “verifiability gap” and modeling reproducibility as an evidence-dependent governance profile. Through a three-phase empirical study comparing locally hosted large models with commercial frontier models, they systematically evaluate how temperature, top_p, top_k, and random seeds affect decision reproducibility, while also analyzing audit latency via execution logs and architectural configurations. Results show that under strict controls, local models achieve 320/320 exact reproductions and hosted models 959/960; however, architectural orchestration significantly influences outcomes, and historical decisions remain difficult to reconstruct—even with deterministic models—demonstrating that verifiability does not automatically improve with model scale.
0 citationsRead paper
Compactly supported radial basis functions as probability density functions
This work addresses the underexplored use of compactly supported radial basis functions (RBFs) in probabilistic density modeling by systematically formulating them as parametric probability density functions. Specifically, it introduces a mixture density model based on the Wendland 𝒞² kernel and derives closed-form expressions for key statistical quantities—including moments and cumulative distribution functions—under both truncated and untruncated support. An incremental learning algorithm is proposed, combining k-means for centroid initialization with stochastic gradient descent for parameter optimization. Experimental results demonstrate that the proposed method achieves likelihood performance and model complexity comparable to Gaussian mixture models on both synthetic and real-world datasets, while enabling exact computation of distributional properties owing to its compact support and analytical tractability.
0 citationsRead paper
An Evaluation of ABR Switching for Time-Shifted Clients in MoQ
This study addresses the vulnerability of MoQ’s default bitrate switching mechanism to playback interruptions under network congestion, which particularly degrades the time-shifted streaming experience. The authors evaluate SWITCH-style adaptive bitrate (ABR) algorithms in a Mininet emulation environment, demonstrating for the first time that standard ABR algorithms—without modification—can effectively support MoQ time-shifted playback. Experimental results reveal a counterintuitive yet significant increase in throughput following rebuffering events, suggesting that existing ABR strategies can directly enhance quality of service for time-shifted scenarios. These findings not only validate the immediate applicability of conventional ABR logic to MoQ-based time-shifted streaming but also highlight key directions for optimizing ABR switching behavior within the MoQ framework.
0 citationsRead paper
UI-LIC: A Unified Framework for Evaluating Learned Image Compression Models
This work addresses the challenge of fair comparison in learned image compression (LIC), which has been hindered by inconsistent model implementations, training protocols, and evaluation metrics. To this end, we present UI-LIC, an open-source unified framework that integrates six state-of-the-art LIC models alongside traditional codecs within a consistent experimental setup, enabling end-to-end automated training, inference, and comparative evaluation. The framework features a graphical user interface supporting bitrate alignment, computation of multiple quality metrics—including PSNR, SSIM, VMAF, and LPIPS—and interactive visualization of quality heatmaps. Deployment and benchmarking require only a single command, substantially lowering the barrier to entry for researchers. The code is publicly released to foster reproducibility and further advancement in the field.
0 citationsRead paper