arXiv subjects

Wei-Jie Xu

Publications and source records attributed to Wei-Jie Xu.

5 recordsLinked to original sources

arXiv2026

How to Achieve the Intended Aim of Deep Clustering Now, without Deep Learning

Deep clustering (DC) is often quoted to have a key advantage over $k$-means clustering. Yet, this advantage is often demonstrated using image datasets only, and it is unclear whether it addresses the fundamental limitations of $k$-means clustering. Deep Embedded Clustering (DEC) learns a latent representation via an autoencoder and performs clustering based on a $k$-means-like procedure, while the optimization is conducted in an end-to-end manner. This paper investigates whether the deep-learned representation has enabled DEC to overcome the known fundamental limitations of $k$-means clustering, i.e., its inability to discover clusters of arbitrary shapes, varied sizes and densities. Our investigations on DEC have a wider implication on deep clustering methods in general. Notably, none of these methods exploit the underlying data distribution. We uncover that a non-deep learning approach achieves the intended aim of deep clustering by making use of distributional information of clusters in a dataset to effectively address these fundamental limitations.

cs.LG↗

arXiv2026

UniCorn: Towards Self-Improving Unified Multimodal Models through Self-Generated Supervision

While Unified Multimodal Models (UMMs) have achieved remarkable success in cross-modal comprehension, a significant gap persists in their ability to leverage such internal knowledge for high-quality generation. We formalize this discrepancy as Conduction Aphasia, a phenomenon where models accurately interpret multimodal inputs but struggle to translate that understanding into faithful and controllable synthesis. To address this, we propose UniCorn, a simple yet elegant self-improvement framework that eliminates the need for external data or teacher supervision. By partitioning a single UMM into three collaborative roles: Proposer, Solver, and Judge, UniCorn generates high-quality interactions via self-play and employs cognitive pattern reconstruction to distill latent understanding into explicit generative signals. To validate the restoration of multimodal coherence, we introduce UniCycle, a cycle-consistency benchmark based on a Text to Image to Text reconstruction loop. Extensive experiments demonstrate that UniCorn achieves comprehensive and substantial improvements over the base model across six general image generation benchmarks. Notably, it achieves SOTA performance on TIIF(73.8), DPG(86.8), CompBench(88.5), and UniCycle while further delivering substantial gains of +5.0 on WISE and +6.5 on OneIG. These results highlight that our method significantly enhances T2I generation while maintaining robust comprehension, demonstrating the scalability of fully self-supervised refinement for unified multimodal intelligence.

cs.CV↗

arXiv2025

PIP: Perturbation-based Iterative Pruning for Large Language Models

The rapid increase in the parameter counts of Large Language Models (LLMs), which often reach into the billions or even trillions, presents significant challenges for their practical deployment, particularly in resource-constrained environments. To address this issue, we propose PIP (Perturbation-based Iterative Pruning), a novel double-view structured pruning method to optimize LLMs, which combines information from two different views: the unperturbed view and the perturbed view. With the calculation of gradient differences, PIP iteratively prunes those that struggle to distinguish between these two views. Our experiments show that PIP reduces the parameter count by approximately 20% while retaining over 85% of the original model's accuracy across varied benchmarks. In some cases, the performance of the pruned model is within 5% of the unpruned version, demonstrating PIP's ability to preserve key aspects of model effectiveness. Moreover, PIP consistently outperforms existing state-of-the-art (SOTA) structured pruning methods, establishing it as a leading technique for optimizing LLMs in constrained environments.

cs.LG↗

arXiv2022

Atomic mass, Bjorken variable and scale dependence of quark transport coefficient in Drell-Yan process for proton incident on nucleus

By means of the nuclear parton distributions determined without the fixed-target Drell-Yan experimental data and the analytic expression of quenching weight based on BDMPS formalism, a next-to-leading order analyses are performed on the Drell-Yan differential cross section ratios from Fermilab E906 and E866 Collaborations. It is found that the calculated results with only the nuclear effects of parton distribution are not in agreement with the E866 and E906 experimental data. The incoming parton energy loss effect can not be ignored in the nuclear Drell-Yan reactions. The predicted results indicate that with the quark transport coefficient as a constant, the suppression due to the target nuclear geometry effect is approximately 19.24% for the quark transport coefficient. It is shown that we should consider the target nuclear geometry effect in studying the Drell-Yan reaction on nuclear targets. On the basis of Bjorken variable and scale dependence of the quark transport coefficient, the atomic mass dependence is incorporated. The quark transport coefficient is determined as a function of the atomic mass, Bjorken variable $x_2$ and scale $Q^2$ by the global fit of the experimental data. The determined constant factor $\hat{q}_0$ of the quark transport coefficient is $0.061\pm0.004 {\rm GeV^2/fm}$. It is found that the atomic mass dependence has a remarkable impact on the constant factor $\hat{q}_0$ in the quark transport coefficient in cold nuclear matter.

hep-ph↗

arXiv2013

Early thermalization of quark-gluon matter with the elastic scattering of ggq and ggqbar

Elastic gluon-gluon-quark (gluon-gluon-antiquark) scattering is studied in perturbative QCD with 123 Feynman diagrams at the tree level. Individually squared amplitudes and interference terms of the Feynman diagrams are derived. With the elastic gluon-gluon-quark scattering and the elastic gluon-gluon-antiquark scattering transport equations are established. In the thermalization process of initially created quark-gluon matter, this matter is governed by elastic 2-to-2 scattering and elastic 3-to-3 scattering. Solutions of the transport equations show that initially created quark-gluon matter takes early thermalization, i.e., thermal states are established rapidly. Different thermalization times of gluon matter and quark matter are obtained.

hep-ph↗