arXiv subjects

Junxi Zhang

Publications and source records attributed to Junxi Zhang.

8 recordsLinked to original sources

arXiv2026

DiffCVE: Diffusion-based Compressed Video Enhancement

Perceptual quality enhancement of severely compressed videos remains challenging due to complex artifact patterns and substantial information loss. Recent diffusion models have demonstrated strong generative capability for visual restoration, but directly applying them to compressed video often ignores compression degradation characteristics and may introduce structure-inconsistent hallucinations. To address this issue, this paper presents a diffusion-based compressed video enhancement method, named DiffCVE. Coding Prior-enhanced Dual Conditioning (CPDC) branches are designed to jointly model compressed video and coding prior conditions, where coding priors including residuals and motion vectors provide complementary structural and motion guidance during the diffusion denoising process. To make the diffusion process aware of compression severity, a Compression Degradation Semantic Prompting (CDSP) mechanism is introduced to leverage QP-conditioned textual prompts together with LoRA fine-tuning. In addition, a Coding Prior-guided Weighted Fusion (CPWF) module is incorporated into the VAE decoder to fuse VAE encoder and coding prior encoder features with QP-predicted weights. Extensive experiments demonstrate the effectiveness of the proposed method in improving perceptual quality, especially under severe compression settings. The project page with enhanced video demonstrations is available at https://wqmaker.github.io/projects/DiffCVE/.

cs.CV↗

arXiv2026

AURA: Adaptive Uncertainty-aware Refinement for LLM-as-a-Judge Auditing

Large language models (LLMs) are increasingly used as judges for open-ended generation, as large-scale human evaluation is often expensive and difficult to scale, yet their preferences remain imperfect proxies for human judgment. Existing auditing pipelines often assume that a reliable subset of examples or clean supervision signals are available beforehand, for example from human annotation, heuristic filtering, or the outputs of strong judges. In LLM evaluation, this assumption is fragile: the initial split may inherit judge bias, while human verification is typically too scarce to define stable groups at scale. We propose AURA, an adaptive uncertainty--aware refinement framework for auditing pairwise LLM--as--a--judge decisions under selected human verification. AURA iteratively learns a human-consistency signal, propagates reliable evidence, and prioritizes uncertain comparisons for human review. The key idea is to treat trust in a judge as a latent quantity that is progressively refined as evidence accumulates. We provide a compact formulation, a stable refinement procedure, and a comprehensive evaluation on both synthetic and real pairwise LLM-answer data.

stat.ML↗

arXiv2024

Large parameter asymptotic analysis for homogeneous normalized random measures with independent increments

Homogeneous normalized random measures with independent increments (hNRMIs) represent a broad class of Bayesian nonparametric priors and thus are widely used. In this paper, we obtain the strong law of large numbers, the central limit theorem and the functional central limit theorem of hNRMIs when the concentration parameter $a$ approaches infinity. To quantify the convergence rate of the obtained central limit theorem, we further study the Berry-Esseen bound, which turns out to be of the form $O \left( \frac{1}{\sqrt{a}}\right)$. As an application of the central limit theorem, we present the functional delta method, which can be employed to obtain the limit of the quantile process of hNRMIs. As an illustration of the central limit theorems, we demonstrate the convergence numerically for the Dirichlet processes and the normalized inverse Gaussian processes with various choices of the concentration parameters.

math.ST↗

arXiv2024

Perceptual-oriented Learned Image Compression with Dynamic Kernel

In this paper, we extend our prior research named DKIC and propose the perceptual-oriented learned image compression method, PO-DKIC. Specifically, DKIC adopts a dynamic kernel-based dynamic residual block group to enhance the transform coding and an asymmetric space-channel context entropy model to facilitate the estimation of gaussian parameters. Based on DKIC, PO-DKIC introduces PatchGAN and LPIPS loss to enhance visual quality. Furthermore, to maximize the overall perceptual quality under a rate constraint, we formulate this challenge into a constrained programming problem and use the Linear Integer Programming method for resolution. The experiments demonstrate that our proposed method can generate realistic images with richer textures and finer details when compared to state-of-the-art image compression techniques.

cs.MM↗

arXiv2024

Managing Temporal Resolution in Continuous Value Estimation: A Fundamental Trade-off

A default assumption in reinforcement learning (RL) and optimal control is that observations arrive at discrete time points on a fixed clock cycle. Yet, many applications involve continuous-time systems where the time discretization, in principle, can be managed. The impact of time discretization on RL methods has not been fully characterized in existing theory, but a more detailed analysis of its effect could reveal opportunities for improving data-efficiency. We address this gap by analyzing Monte-Carlo policy evaluation for LQR systems and uncover a fundamental trade-off between approximation and statistical error in value estimation. Importantly, these two errors behave differently to time discretization, leading to an optimal choice of temporal resolution for a given data budget. These findings show that managing the temporal resolution can provably improve policy evaluation efficiency in LQR systems with finite data. Empirically, we demonstrate the trade-off in numerical simulations of LQR instances and standard RL benchmarks for non-linear continuous control.

cs.LG↗

arXiv2023

Large sample asymptotic analysis for normalized random measures with independent increments

Normalized random measures with independent increments represent a large class of Bayesian nonaprametric priors and are widely used in the Bayesian nonparametric framework. In this paper, we provide the posterior consistency analysis for normalized random measures with independent increments (NRMIs) through the corresponding Levy intensities used to characterize the completely random measures in the construction of NRMIs. Assumptions are introduced on the Levy intensities to analyze the posterior consistency of NRMIs and are verified with multiple interesting examples. A focus of the paper is the Bernstein-von Mises theorem for the normalized generalized gamma process (NGGP) when the true distribution of the sample is discrete or continuous. When the Bernstein-von Mises theorem is applied to construct credible sets, in addition to the usual form there will be an additional bias term on the left endpoint closely related to the number of atoms of the true distribution when it is discrete. We also discuss the affect of the estimators for the model parameters of the NGGP under the Bernstein-von Mises convergences. Finally, to further explain the necessity of adding the bias correction in constructing credible sets, we illustrate numerically how the bias correction affects the coverage of the true value by the credible sets when the true distribution is discrete.

math.ST↗

arXiv2022

Efficient and Accurate Quantized Image Super-Resolution on Mobile NPUs, Mobile AI & AIM 2022 challenge: Report

Image super-resolution is a common task on mobile and IoT devices, where one often needs to upscale and enhance low-resolution images and video frames. While numerous solutions have been proposed for this problem in the past, they are usually not compatible with low-power mobile NPUs having many computational and memory constraints. In this Mobile AI challenge, we address this problem and propose the participants to design an efficient quantized image super-resolution solution that can demonstrate a real-time performance on mobile NPUs. The participants were provided with the DIV2K dataset and trained INT8 models to do a high-quality 3X image upscaling. The runtime of all models was evaluated on the Synaptics VS680 Smart Home board with a dedicated edge NPU capable of accelerating quantized neural networks. All proposed solutions are fully compatible with the above NPU, demonstrating an up to 60 FPS rate when reconstructing Full HD resolution images. A detailed description of all models developed in the challenge is provided in this paper.

eess.IV↗

arXiv2020

Functional central limit theorems for stick-breaking priors

We obtain the empirical strong law of large numbers, empirical Glivenko-Cantelli theorem, central limit theorem, functional central limit theorem for various nonparametric Bayesian priors which include the Dirichlet process with general stick-breaking weights, the Poisson-Dirichlet process, the normalized inverse Gaussian process, the normalized generalized gamma process, and the generalized Dirichlet process. For the Dirichlet process with general stick-breaking weights, we introduce two general conditions such that the central limit theorem and functional central limit theorem hold. Except in the case of the generalized Dirichlet process, since the finite dimensional distributions of these processes are either hard to obtain or are complicated to use even they are available, we use the method of moments to obtain the convergence results. For the generalized Dirichlet process we use its finite dimensional marginal distributions to obtain the asymptotics although the computations are highly technical.

math.ST↗