arXiv subjects

Jiaxing Zhao

Publications and source records attributed to Jiaxing Zhao.

At least 55 records · Page 3Linked to original sources

arXiv2025

R1-Omni: Explainable Omni-Multimodal Emotion Recognition with Reinforcement Learning

In this work, we present the first application of Reinforcement Learning with Verifiable Reward (RLVR) to an Omni-multimodal large language model in the context of emotion recognition, a task where both visual and audio modalities play crucial roles. We leverage RLVR to optimize the Omni model, significantly enhancing its performance in three key aspects: reasoning capability, emotion recognition accuracy, and generalization ability. The introduction of RLVR not only improves the model's overall performance on in-distribution data but also demonstrates superior robustness when evaluated on out-of-distribution datasets. More importantly, the improved reasoning capability enables clear analysis of the contributions of different modalities, particularly visual and audio information, in the emotion recognition process. This provides valuable insights into the optimization of multimodal large language models.

cs.LG↗

arXiv2025

Hadronization of Heavy Quarks

Heavy-flavor hadrons produced in ultra-relativistic heavy-ion collisions are a sensitive probe for studying hadronization mechanisms of the quark-gluon-plasma. In this work, we survey how different transport models for the simulation of heavy-quark diffusion through a quark-gluon plasma in heavy-ion collisions implement hadronization and how this affects final-state observables. Utilizing the same input charm-quark distribution in all models at the hadronization transition, we find that the transverse-momentum dependence of the nuclear modification factor of various charm hadron species has significant sensitivity to the hadronization scheme. In addition, the charm-hadron elliptic flow exhibits a nontrivial dependence on the elliptic flow of the hadronizing partonic medium.

hep-ph↗

arXiv2025

Heavy quark potential and thermal charm production in heavy-ion collisions

Heavy quark mass in QGP is related to the heavy quark potential at a large distance. In this study we test three different heavy quark potentials, namely, the free energy, the internal energy of the heavy quark pair in QGP, and the unscreened potential, which was recently proposed by the HotQCD Collaboration, through the thermal production of charm quarks in heavy-ion collisions at the LHC. We find that the free energy potential overestimates charm production in heavy-ion collisions at the LHC, while the unscreened potential produces results closest to the experimental data from the ALICE Collaboration among the three potentials.

nucl-th↗

arXiv2025

HumanOmni: A Large Vision-Speech Language Model for Human-Centric Video Understanding

In human-centric scenes, the ability to simultaneously understand visual and auditory information is crucial. While recent omni models can process multiple modalities, they generally lack effectiveness in human-centric scenes due to the absence of large-scale, specialized datasets and non-targeted architectures. In this work, we developed HumanOmni, the industry's first human-centric Omni-multimodal large language model. We constructed a dataset containing over 2.4 million human-centric video clips with detailed captions and more than 14 million instructions, facilitating the understanding of diverse human-centric scenes. HumanOmni includes three specialized branches for understanding different types of scenes. It adaptively fuses features from these branches based on user instructions, significantly enhancing visual understanding in scenes centered around individuals. Moreover, HumanOmni integrates audio features to ensure a comprehensive understanding of environments and individuals. Our experiments validate HumanOmni's advanced capabilities in handling human-centric scenes across a variety of tasks, including emotion recognition, facial expression description, and action understanding. Our model will be open-sourced to facilitate further development and collaboration within both academia and industry.

cs.CV↗

arXiv2025

Facial Dynamics in Video: Instruction Tuning for Improved Facial Expression Perception and Contextual Awareness

Facial expression captioning has found widespread application across various domains. Recently, the emergence of video Multimodal Large Language Models (MLLMs) has shown promise in general video understanding tasks. However, describing facial expressions within videos poses two major challenges for these models: (1) the lack of adequate datasets and benchmarks, and (2) the limited visual token capacity of video MLLMs. To address these issues, this paper introduces a new instruction-following dataset tailored for dynamic facial expression caption. The dataset comprises 5,033 high-quality video clips annotated manually, containing over 700,000 tokens. Its purpose is to improve the capability of video MLLMs to discern subtle facial nuances. Furthermore, we propose FaceTrack-MM, which leverages a limited number of tokens to encode the main character's face. This model demonstrates superior performance in tracking faces and focusing on the facial expressions of the main characters, even in intricate multi-person scenarios. Additionally, we introduce a novel evaluation metric combining event extraction, relation classification, and the longest common subsequence (LCS) algorithm to assess the content consistency and temporal sequence consistency of generated text. Moreover, we present FEC-Bench, a benchmark designed to assess the performance of existing video MLLMs in this specific task. All data and source code will be made publicly available.

cs.CV↗

arXiv2025

Learning Hadron Emitting Sources with Deep Neural Networks

The correlation function observed in high-energy collision experiments encodes critical information about the emitted source and hadronic interactions. While the proton-proton interaction potential is well constrained by nucleon-nucleon scattering data, these measurements offer a unique avenue to investigate the proton-emitting source, reflecting the dynamical properties of the collisions. In this Letter, we present an unbiased approach to reconstruct proton-emitting sources from experimental correlation functions. Within an automatic differentiation framework, we parameterize the source functions with deep neural networks, to compute correlation functions. This approach achieves a lower chi-squared value compared to conventional Gaussian source functions and captures the long-tail behavior, in qualitative agreement with simulation predictions.

nucl-th↗

arXiv2024

Charmonia Production in Hot QCD Matter and Electromagnetic Fields

Both hot QCD matter and extremely strong electromagnetic fields are generated in relativistic heavy-ion collisions. We employ the transport model and the equivalent photon approximation (EPA) to study charmonium hadroproduction and photoproduction in nucleus-nucleus collisions, respectively. In photoproduction, quasi-real photons may interact with the whole nucleus or individual nucleons, which is called the coherent and incoherent processes, respectively. The typical momentum of charmonium produced in two processes is located in $p_T\lesssim 1/R_A$ and $p_T\lesssim 1/R_N$, where $R_A$ and $R_N$ are the radii of nucleus and the nucleon. Both kinds of photoproduction and also hadroproduction are considered to calculate charmonium production in different transverse momentum bins, rapidity bins, and collision centralities, incorporating modifications from hot QCD matter and initial cold nuclear matter effects. Our calculations explain experimental data about charmonium nuclear modification factors and the production cross-section in ultra-peripheral collisions. Charmonium nuclear modification is far above the unit at extremely low $p_T$ ($p_T < 0.1$ GeV/c) in peripheral collisions with centrality 70-90\%, attributed to coherent photoproduction.

nucl-th↗

arXiv2024

Recent theoretical developments in quarkonia production in relativistic heavy ion collisions

Quarkonia are golden probes of heavy ion collisions that have attracted much attention from both experimental and theoretical perspectives. This paper will review recent theoretical studies on quarkonium thermal properties, with a particular focus on the heavy quark finite-temperature potential obtained by Lattice QCD and other approaches. Moreover, it will examine the advancements in the real-time evolution of quarkonia in heavy ion collisions.

hep-ph↗

arXiv2024

Cluster and anti-cluster production in heavy-ion collisions and pA reactions

We investigate light cluster and anti-cluster production in heavy-ion collisions from SIS to RHIC energies within the Parton-Hadron-Quantum-Molecular Dynamics (PHQMD) microscopic transport approach which propagates (anti-)baryons using n-body QMD dynamics. In PHQMD the clusters are formed dynamically by potential interactions between baryons - and recognized by the Minimum Spanning Tree (MST) algorithm - as well as by kinetic reactions in case of deuterons. We present the novel PHQMD results for different observables such as excitation functions of the multiplicity of deuterons, anti-deuterons and tritons, as well as their transverse momentum spectra. Moreover, we investigate the system size dependence of proton and deuteron production in p+A collisions and show the PHQMD results for p+A collisions (A = Be, Al, Cu, Au) at 14 AGeV/c, as well as for asymmetric Au+A collisions (A = Al, Cu, Pb) at a bombarding energy of about 10 AGeV.

nucl-th↗

arXiv2024

Quarkonium production in pp and heavy-ion collisions

We describe bottomonium production not only in pp collisions but also in heavy-ion collisions by using the Remler's formalism where quarkonium density operator is applied to all possible combination of heavy quark and heavy antiquark pairs. In pp collisions heavy (anti)quark momentum is provided by the PYTHIA event generator after rescaling $p_T$ and rapidity to imitate the FONLL calculations. Then spatial separation between heavy quark and heavy antiquark is introduced based on the uncertainty principle. In heavy-ion collisions quarkonium wavefunction changes with temperature assuming heavy quark potential equals the free energy of heavy quark and heavy antiquark system in heat bath. The density operator is updated whenever heavy quark or heavy antiquark scatters in QGP produced in heavy-ion collisions. Our results are consistent with the experimental data from ALICE and CMS Collaborations assuming that the interaction rate of heavy (anti)quark in quarkonium is suppressed to 10 \% that of unbound heavy (anti)quark. We also find that off-diagonal recombination of bottomonium barely happens even in Pb+Pb collisions at $\sqrt{s}=5.02$ TeV.

hep-ph↗

arXiv2024

Gluon decay into heavy quark pair under a strong magnetic field

Due to the extreme large magnetic field produced in the initial stage of non-central heavy-ion collision, the dynamical process of gluon decay into heavy quark pair will take place under an external field rather than in vacuum. Unlike in the vacuum case, where the process is forbidden by energy momentum conservation, under the external field, a process emerges considering the background energy which recovers the conservation. We calculate the gluon decay rate at leading order under a uniform magnetic field.

hep-ph↗

arXiv2024

$B_c$ enhancement with non-thermalized bottom quarks in nuclear collisions at Large Hadron Collider

We study $B_c$ production in high-energy nuclear collisions in a transport approach with dissociation and regeneration at finite temperatures. Due to the rare production in p+p collisions and the strong combination of uncorrelated $c$ and $\bar b$ quarks in the quark-gluon plasma, the $B_c$ yield is significantly enhanced in the nuclear collisions at the Large Hadron Collider. Moreover, the centrality and momentum-dependent yield of $B_c$ sensitively reflect the thermalization degree of bottom quarks. And the newly observed experimental data favors a far from the thermal bottom quark distribution in the quark-gluon plasma.

hep-ph↗

arXiv2024

System size dependence of energy loss and correlations of heavy mesons at LHC energies

We study the system size dependence of heavy quark (HQ) observables at a center of mass energy of $\sqrt{s_{\rm NN}}=5.02$ TeV to explore whether it can provide further constraints on the physical processes which are involved: energy loss of HQs in the quark gluon plasma (QGP), hadronization and hadronic rescattering. We use the EPOS4HQ approach to study p-p and 0-10\% central O-O, Ar-Ar, Kr-Kr and Pb-Pb reactions and investigate in detail the momentum change of heavy quarks from creation until their detection as part of a hadron as well as the enhancement of heavy baryon production at low $p_T$. We investigate furthermore the origin and the system size dependence of the azimuthal correlations between the heavy quark $Q$ and the heavy antiquark $\bar Q$ and how one can bypass the problem that correlations are washed out due to the combinatorial background. We conclude that a systematic study of the system size dependence of the momentum loss allows to separate the momentum loss due to the passage through the QGP from the momentum change due to hadronization and the $Q\bar Q$ correlations allow to gain inside into the different pQCD processes in which the heavy quarks are created.

hep-ph↗

arXiv2024

Production of doubly charmed hadron $Ξ_{cc}^{++}$ and $T_{cc}^+$ in relativistic heavy ion collisions

Heavy ion collisions provide a unique opportunity for studying the properties of exotic hadrons with two charm quarks. The production of $T_{cc}^+$ is significantly enhanced in nuclear collisions compared to proton-proton collisions due to the creation of multiple charm pairs. In this study, we employ the Langevin equation in combination with the Instantaneous Coalescence Model (LICM) to investigate the production of $T_{cc}^+$ and $Ξ_{cc}^{++}$ which consists of two charm quarks. We consider $T_{cc}^+$ as molecular states composed of $D$ and $D^*$ mesons. The Langevin equation is used to calculate the energy loss of charm quarks and $D$ mesons in the hot medium. The hadronization process, where charm quarks transform into each $D$ state as constituents of $T_{cc}^+$ production, is described using the coalescence model. The coalescence probability between $D$ and $D^*$ is determined by the Wigner function, which encodes the information of the $T_{cc}^+$ wave function. Our results show that the $T_{cc}^+$ production varies by approximately one order of magnitude when different widths in the Wigner function, representing distinct binding energies of $T_{cc}^+$, are considered. This variation offers valuable insights into the nature of $T_{cc}^+$ through the analysis of its wave function. The $Ξ_{cc}^{++}$ is treated as a hadronic state produced at the hadronization of the deconfined matter. Its production is also calculated as a comparison with the molecular state $T_{cc}^+$.

nucl-th↗

arXiv2024

D-Aug: Enhancing Data Augmentation for Dynamic LiDAR Scenes

Creating large LiDAR datasets with pixel-level labeling poses significant challenges. While numerous data augmentation methods have been developed to reduce the reliance on manual labeling, these methods predominantly focus on static scenes and they overlook the importance of data augmentation for dynamic scenes, which is critical for autonomous driving. To address this issue, we propose D-Aug, a LiDAR data augmentation method tailored for augmenting dynamic scenes. D-Aug extracts objects and inserts them into dynamic scenes, considering the continuity of these objects across consecutive frames. For seamless insertion into dynamic scenes, we propose a reference-guided method that involves dynamic collision detection and rotation alignment. Additionally, we present a pixel-level road identification strategy to efficiently determine suitable insertion positions. We validated our method using the nuScenes dataset with various 3D detection and tracking methods. Comparative experiments demonstrate the superiority of D-Aug.

cs.CV↗

arXiv2024

A numerical algorithm for solving the coupled Schrödinger equations using inverse power method

The inverse power method is a numerical algorithm to obtain the eigenvectors of a matrix. In this work, we develop an iteration algorithm, based on the inverse power method, to numerically solve the Schrödinger equation that couples an arbitrary number of components. Such an algorithm can also be applied to the multi-body systems. To show the power and accuracy of this method, we also present an example of solving the Dirac equation under the presence of an external scalar potential and a constant magnetic field, with source code publicly available.

physics.comp-ph↗

arXiv2024

Heavy flavour hadron production in relativistic heavy ion collisions at RHIC and LHC in EPOS4HQ

Employing the recently developed EPOS4HQ event generator, we study the production of different heavy-flavor mesons in relativistic heavy-ion collisions at RHIC and LHC energies. The transverse momentum spectra, yield ratio, nuclear modification factor, and elliptic flow can be well described in the EPOS4HQ framework. We furthermore analyze the processes which modify these observables as compared to $pp$ collisions and are at the origin of the experimentally determined nuclear modification factor $R_{AA}$.

hep-ph↗

arXiv2024

Bottomonium evolution with in-medium heavy quark potential from lattice QCD

The static properties and dynamic evolution of bottomonium states in a hot QCD medium are investigated through the Schrödinger equation with complex heavy quark potentials, which are presented recently in lattice QCD study and with three different extractions. This approach builds a direct connection between the in-medium heavy quark potentials from the lattice QCD to the experimental observables. The yields and nuclear modification factors $R_{AA}$ of bottomonium in Pb-Pb collisions at $\sqrt{s_{\rm NN}}=5.02~\rm TeV$ are calculated in this work. Our results show a large suppression of the bottomonium yield in heavy ion collisions due to the large imaginary potential. To understand bottomonium $R_{AA}$ based on lattice QCD potentials, we propose a fomration time for bottomonium states and find that experimental data can be well explained with the heavy quark potential extracted by the Padé fit, which shows no color screening in the real part potential.

nucl-th↗