arXiv subjects

Yayun He

Publications and source records attributed to Yayun He.

At least 19 recordsLinked to original sources

arXiv2026

ProbPlug: A Plugin Uncertainty Network for Reliable Confidence in LLM Binary Classification

Large language models (LLMs) have achieved strong performance across a broad range of classification settings, yet the reliability of their predictions remains a major obstacle to deployment in high-stakes scenarios. Although confidence estimation for LLMs has been widely studied, confidence calibration for LLM-based classification remains underexplored. We introduce ProbPlug, a lightweight confidence estimation framework for LLM-based binary classification, which predicts whether an output is correct using internal token features extracted from a frozen LLM. ProbPlug employs a self-attention module to aggregate hidden representations and can be integrated into the original inference pipeline without modifying the base model. Experiments across multiple tasks involving both text-based and multimodal large models show that ProbPlug provides more reliable confidence estimates, improves classification performance with negligible additional overhead, and exhibits strong generalization across tasks. These results indicate that ProbPlug serves as a practical solution for confidence estimation in LLM-based classification. Our code is publicly available at Github.

cs.CL↗

arXiv2026

Evolvable Embodied Agent for Robotic Manipulation via Long Short-Term Reflection and Optimization

Achieving general-purpose robotics requires empowering robots to adapt and evolve based on their environment and feedback. Traditional methods face limitations such as extensive training requirements, difficulties in cross-task generalization, and lack of interpretability. Prompt learning offers new opportunities for self-evolving robots without extensive training, but simply reflecting on past experiences. However, extracting meaningful insights from task successes and failures remains a challenge. To this end, we propose the evolvable embodied agent (EEAgent) framework, which leverages large vision-language models (VLMs) for better environmental interpretation and policy planning. To enhance reflection on past experiences, we propose a long short-term reflective optimization (LSTRO) mechanism that dynamically refines prompts based on both past experiences and newly learned lessons, facilitating continuous self-evolution, thereby enhancing overall task success rates. Evaluations on six VIMA-Bench tasks reveal that our approach sets a new state-of-the-art, notably outperforming baselines in complex scenarios.

cs.RO↗

arXiv2026

VLA-InfoEntropy: A Training-Free Vision-Attention Information Entropy Approach for Vision-Language-Action Models Inference Acceleration and Success

Vision-Language-Action (VLA) models integrate visual perception, language understanding, and action decision-making for cross-modal semantic alignment, exhibiting broad application potential. However, the joint processing of high-dimensional visual features, complex linguistic inputs, and continuous action sequences incurs significant computational overhead and low inference efficiency, thereby hindering real-time deployment and reliability. To address this issue, we use image entropy to quantify the grayscale distribution characteristics of each visual token and introduce attention entropy to capture the distribution of attention scores over task-related text. Visual entropy identifies texture-rich or structurally informative regions, while attention entropy pinpoints semantically relevant tokens. Combined with timestep information, these metrics enable a dynamic transition strategy that shifts the model's focus from global visual features to attention-guided local informative regions. Thus, the resulting VLA-InfoEntropy method integrates spatial, semantic, and temporal cues to reduce redundancy while preserving critical content. Extensive experiments show that our method reduces inference parameters, accelerates inference speed, and outperforms existing approaches.

cs.CV↗

arXiv2026

Confusion-Aware In-Context-Learning for Vision-Language Models in Robotic Manipulation

Vision-language models (VLMs) have significantly improved the generalization capabilities of robotic manipulation. However, VLM-based systems often suffer from a lack of robustness, leading to unpredictable errors, particularly in scenarios involving confusable objects. Our preliminary analysis reveals that these failures are mainly caused by shortcut learning problem inherently in VLMs, limiting their ability to accurately distinguish between confusable features. To this end, we propose Confusion-Aware In-Context Learning (CAICL), a method that enhances VLM performance in confusable scenarios for robotic manipulation. The approach begins with confusion localization and analysis, identifying potential sources of confusion. This information is then used as a prompt for the VLM to focus on features most likely to cause misidentification. Extensive experiments on the VIMA-Bench show that CAICL effectively addresses the shortcut learning issue, achieving a 85.5\% success rate and showing good stability across tasks with different degrees of generalization.

cs.RO↗

arXiv2026

Attention-weighted Centered Kernel Alignment for Knowledge Distillation in Large Audio-Language Models Applied to Speech Emotion Recognition

The emergence of Large Audio-Language Models (LALMs) has advanced Speech Emotion Recognition (SER), but their size limits deployment in resource-constrained environments. While Knowledge Distillation is effective for LALM compression, existing methods remain underexplored in distilling the cross-modal projection module (Projector), and often struggle with alignment due to differences in feature dimensions. We propose PL-Distill, a KD framework that combines Projector-Level Distillation (PDist) to align audio embeddings and Logits-Level Distillation (LDist) to align output logits. PDist introduces Attention-weighted Centered Kernel Alignment, a novel approach we propose to highlight important time steps and address dimension mismatches. Meanwhile, LDist minimizes the Kullback-Leibler divergence between teacher and student logits from audio and text modalities. On IEMOCAP, RAVDESS, and SAVEE, PL-Distill compresses an 8.4B-parameter teacher to a compact 1.1B-parameter student, consistently outperforming the teacher, state-of-the-art pretrained models, and other KD baselines across all metrics.

cs.SD↗

arXiv2025

EMO-RL: Emotion-Rule-Based Reinforcement Learning Enhanced Audio-Language Model for Generalized Speech Emotion Recognition

Although Large Audio-Language Models (LALMs) have exhibited outstanding performance in auditory understanding, their performance in affective computing scenarios, particularly in emotion recognition, reasoning, and subtle sentiment differentiation, remains suboptimal. Recent advances in Reinforcement Learning (RL) have shown promise in improving LALMs' reasoning abilities. However, two critical challenges hinder the direct application of RL techniques to Speech Emotion Recognition (SER) tasks: (1) convergence instability caused by ambiguous emotional boundaries and (2) limited reasoning ability when using relatively small models (e.g., 7B-parameter architectures). To overcome these limitations, we introduce EMO-RL, a novel framework incorporating reinforcement learning with two key innovations: Emotion Similarity-Weighted Reward (ESWR) and Explicit Structured Reasoning (ESR). Built upon pretrained LALMs, our method employs group-relative policy optimization with emotion constraints. Comprehensive experiments demonstrate that our EMO-RL training strategies can significantly enhance the emotional reasoning capabilities of LALMs, attaining state-of-the-art results on both the MELD and IEMOCAP datasets, and cross-dataset experiments prove the strong superiority of generalization.

cs.SD↗

arXiv2025

Interplay of prompt and non-prompt photons in photon-triggered jet observables

Prompt photons are important yet challenging to observe in relativistic heavy-ion collisions, as they are produced in the early stages and traverse almost the entire QGP medium without interaction. Experimental analyses typically employ isolation cuts, in the hope to identify prompt photons. Most theoretical studies consider only events with actual prompt photons, assuming no contribution from isolated non-prompt photons to reduce computational cost. For the first time, we present a study that compares simulation results generated using inclusive (bremsstrahlung) and prompt-photon events with multiple experimental observables for both $p-p$ and $Pb-Pb$ collisions at $5.02$ TeV. Simulations are carried out using the multi-stage JETSCAPE framework tuned to describe the quenching of jets and hadrons. Isolated non-prompt photons are generated in hard photon bremsstrahlung, where the photon is radiated at a sufficient angle to the jet. Several photon triggered jet and jet substructure observables show significant contributions from inclusive photons, yielding an improvement in comparison with experimental data. Novel photon triggered jet substructure observables are also expected to show new structures, yet to be detected in experiment. This effort examines the significance of isolated non-prompt photons using parameters tuned for a simultaneous description of the leading hadron and jet spectrum, and thus provides an independent verification of the multistage evolution framework.

hep-ph↗

arXiv2025

Effects of hadronic reinteraction on jet fragmentation from small to large systems

We investigate the impact of the hadronic phase on jet quenching in nuclear collider experiments, an open question in heavy-ion physics. Previous studies in a simplified setup suggest that hadronic interactions could have significant effects, but a systematic analysis is needed. Using the X-SCAPE event generator with the SMASH afterburner, we study the role of hadronic rescattering on jet fragmentation hadrons. Applying this framework to $e^++e^-$ collisions, we demonstrate that even in small systems with limited particle production, hadronic interactions lead to measurable modifications in final-state hadronic and jet observables by comparing scenarios with and without afterburner rescattering.

hep-ph↗

arXiv2025

Generalized Audio Deepfake Detection Using Frame-level Latent Information Entropy

Generalizability, the capacity of a robust model to perform effectively on unseen data, is crucial for audio deepfake detection due to the rapid evolution of text-to-speech (TTS) and voice conversion (VC) technologies. A promising approach to differentiate between bonafide and spoof samples lies in identifying intrinsic disparities to enhance model generalizability. From an information-theoretic perspective, we hypothesize the information content is one of the intrinsic differences: bonafide sample represents a dense, information-rich sampling of the real world, whereas spoof sample is typically derived from lower-dimensional, less informative representations. To implement this, we introduce frame-level latent information entropy detector(f-InfoED), a framework that extracts distinctive information entropy from latent representations at the frame level to identify audio deepfakes. Furthermore, we present AdaLAM, which extends large pre-trained audio models with trainable adapters for enhanced feature extraction. To facilitate comprehensive evaluation, the audio deepfake forensics 2024 (ADFF 2024) dataset was built by the latest TTS and VC methods. Extensive experiments demonstrate that our proposed approach achieves state-of-the-art performance and exhibits remarkable generalization capabilities. Further analytical studies confirms the efficacy of AdaLAM in extracting discriminative audio features and f-InfoED in leveraging latent entropy information for more generalized deepfake detection.

cs.SD↗

arXiv2024

Efficient Multi-Model Fusion with Adversarial Complementary Representation Learning

Single-model systems often suffer from deficiencies in tasks such as speaker verification (SV) and image classification, relying heavily on partial prior knowledge during decision-making, resulting in suboptimal performance. Although multi-model fusion (MMF) can mitigate some of these issues, redundancy in learned representations may limits improvements. To this end, we propose an adversarial complementary representation learning (ACoRL) framework that enables newly trained models to avoid previously acquired knowledge, allowing each individual component model to learn maximally distinct, complementary representations. We make three detailed explanations of why this works and experimental results demonstrate that our method more efficiently improves performance compared to traditional MMF. Furthermore, attribution analysis validates the model trained under ACoRL acquires more complementary knowledge, highlighting the efficacy of our approach in enhancing efficiency and robustness across tasks.

cs.LG↗

arXiv2024

Retrieval-Augmented Audio Deepfake Detection

With recent advances in speech synthesis including text-to-speech (TTS) and voice conversion (VC) systems enabling the generation of ultra-realistic audio deepfakes, there is growing concern about their potential misuse. However, most deepfake (DF) detection methods rely solely on the fuzzy knowledge learned by a single model, resulting in performance bottlenecks and transparency issues. Inspired by retrieval-augmented generation (RAG), we propose a retrieval-augmented detection (RAD) framework that augments test samples with similar retrieved samples for enhanced detection. We also extend the multi-fusion attentive classifier to integrate it with our proposed RAD framework. Extensive experiments show the superior performance of the proposed RAD framework over baseline methods, achieving state-of-the-art results on the ASVspoof 2021 DF set and competitive results on the 2019 and 2021 LA sets. Further sample analysis indicates that the retriever consistently retrieves samples mostly from the same speaker with acoustic characteristics highly consistent with the query audio, thereby improving detection performance.

cs.SD↗

arXiv2024

Asymmetric jet shapes with 2D jet tomography

Two-dimensional (2D) jet tomography is a promising tool to study jet medium modification in high-energy heavy-ion collisions. It combines gradient (transverse) and longitudinal jet tomography for selection of events with localized initial jet production positions. It exploits the transverse asymmetry and energy loss that depend, respectively, on the transverse gradient and jet path length inside the quark-gluon plasma (QGP). In this study, we employ the 2D jet tomography to study medium modification of the jet shape of $\gamma$-triggered jets within the linear Boltzmann transport (LBT) model for jet propagation in heavy-ion collisions. Our results show that jets with small transverse asymmetry ($A_N^{\vec{n}}$) or small $\gamma$-jet asymmetry ($x_{J\gamma}=p_T^{\rm jet}/p_T^\gamma$) exhibit a broader jet shape than those with larger $A_N^{\vec{n}}$ or $x_{J\gamma}$, since the former are produced at the center and go through longer path lengths while the later are off-center and close to the surface of the QGP fireball. In events with finite values of $A_N^{\vec{n}}$, jet shapes are asymmetric with respect to the event plane. Hard partons at the core of the jet are deflected away from the denser region while soft partons from the medium response at large angles flow toward the denser part of QGP. Future experimental measurements of these asymmetric features of the jet shape can be used to study the transport properties of jets and medium responses.

hep-ph↗

arXiv2023

3D Multi-system Bayesian Calibration with Energy Conservation to Study Rapidity-dependent Dynamics of Nuclear Collisions

Considerable information about the early-stage dynamics of heavy-ion collisions is encoded in the rapidity dependence of measurements. To leverage the large amount of experimental data, we perform a systematic analysis using three-dimensional hydrodynamic simulations of multiple collision systems -- large and small, symmetric and asymmetric. Specifically, we perform fully 3D multi-stage hydrodynamic simulations initialized by a parameterized model for rapidity-dependent energy deposition, which we calibrate on the hadron multiplicity and anisotropic flow coefficients. We utilize Bayesian inference to constrain properties of the early- and late- time dynamics of the system, and highlight the impact of enforcing global energy conservation in our 3D model.

nucl-th↗

arXiv2023

Hybrid Hadronization of Jet Showers from $e^++e^-$ to $A+A$ with JETSCAPE

In this talk we review jet production in a large variety of collision systems using the JETSCAPE event generator and Hybrid Hadronization. Hybrid Hadronization combines quark recombination, applicable when distances between partons in phase space are small, and string fragmentation appropriate for dilute parton systems. It can therefore smoothly describe the transition from very dilute parton systems like $e^++e^-$ to full $A+A$ collisions. We test this picture by using JETSCAPE to generate jets in various systems. Comparison to experimental data in $e^++e^-$ and $p+p$ collisions allows for a precise tuning of vacuum baseline parameters in JETSCAPE and Hybrid Hadronization. Proceeding to systems with jets embedded in a medium, we study in-medium hadronization for jet showers. We quantify the effects of an ambient medium, focusing in particular on the dependence on the collective flow and size of the medium. Our results clarify the effects we expect from in-medium hadronization of jets on observables like fragmentation functions, hadron chemistry and jet shape.

hep-ph↗

arXiv2023

A multistage framework for studying the evolution of jets and high-$p_T$ probes in small collision systems

Understanding the modification of jets and high-$p_T$ probes in small systems requires the integration of soft and hard physics. We present recent developments in extending the JETSCAPE framework to build an event generator, which includes correlations between soft and hard partons, to study jet observables in small systems. The multi-scale physics of the collision is separated into different stages. Hard scatterings are first sampled at binary collision positions provided by the Glauber geometry. They are then propagated backward in space-time following an initial-state shower to obtain the initiating partons' energies and momenta before the collision. These energies and momenta are then subtracted from the incoming colliding nucleons for soft-particle production, modeled by the 3D-Glauber + hydrodynamics + hadronic transport framework. This new hybrid approach (X-SCAPE) includes non-trivial correlations between jet and soft particle productions in small systems. We calibrate this framework with the final state hadrons' $p_T$-spectra from low to high $p_T$ in $p$-$p$, and and then compare with the spectra in $p$-$Pb$ collisions from the LHC. We also present results for additional observables such as the distributions of event activity as a function of the hardest jet $p_T$ in forward and mid-rapidity for both $p$-$p$ and $p$-$Pb$ collisions.

hep-ph↗

arXiv2023

VoiceExtender: Short-utterance Text-independent Speaker Verification with Guided Diffusion Model

Speaker verification (SV) performance deteriorates as utterances become shorter. To this end, we propose a new architecture called VoiceExtender which provides a promising solution for improving SV performance when handling short-duration speech signals. We use two guided diffusion models, the built-in and the external speaker embedding (SE) guided diffusion model, both of which utilize a diffusion model-based sample generator that leverages SE guidance to augment the speech features based on a short utterance. Extensive experimental results on the VoxCeleb1 dataset show that our method outperforms the baseline, with relative improvements in equal error rate (EER) of 46.1%, 35.7%, 10.4%, and 5.7% for the short utterance conditions of 0.5, 1.0, 1.5, and 2.0 seconds, respectively.

cs.SD↗

arXiv2023

Probing the Short-Distance Structure of the Quark-Gluon Plasma with Energy Correlators

Energy-energy-correlators (EEC's) are a promising observable to study the dynamics of jet evolution in the quark-gluon plasma (QGP) through its imprint on angular scales in the energy flux of final-state particles. We carry out the first complete calculation of EEC's using realistic simulations of high-energy heavy-ion collisions, and dissect the different dynamics underlying the final distribution through analyses of jet propagation in a uniform medium. The EEC's of $\gamma$-jets in heavy-ion collisions are found to be enhanced by the medium response from elastic scatterings instead of induced gluon radiation at large angles. In the meantime, EEC's are suppressed at small angles due to energy loss and transverse momentum broadening of jet shower partons. These modifications are further shown to be sensitive to the angular scale of the in-medium interaction, as characterized by the Debye screening mass. Experimental verification and measurement of such modifications will shed light on this scale, and the short-distance structure of the QGP in heavy-ion collisions.

hep-ph↗

arXiv2023

Linear Boltzmann transport for jet propagation in the quark-gluon plasma: Inelastic processes and jet modification

A Linear Boltzmann Transport (LBT) Monte Carlo model has been developed to describe jet propagation and interaction with the quark-gluon plasma (QGP) in relativistic heavy-ion collisions. A complete set of elastic scattering processes and medium-induced gluon emissions based on the higher-twist formalism are incorporated for both jet shower and medium recoil partons. It has been employed to describe experimental data on large transverse momentum hadron and jet spectra, correlation and jet substructures in high-energy heavy-ion collisions. We document in detail the structure of the model and validation of the Monte Carlo implementations of the physics processes in LBT, in particular, the inelastic process of medium-induced gluon radiation. We carry out a comprehensive examination of the jet-medium interaction as implemented in LBT through energy loss and momentum broadening of a single hard parton, the energy and transverse momentum transfer from leading partons to medium-induced gluons and jet-induced medium excitation, and medium modification of reconstructed jets in a static and uniform medium. With realistic and event-by-event hydrodynamic medium in heavy-ion collisions, we compute and compare to experimental data on the jet cone-size dependence of the single inclusive jet suppression at both Relativistic Heavy-Ion Collider (RHIC) and the Large Hadron Collider (LHC), the dijet asymmetry at LHC and $\gamma$-jet correlation at RHIC. Effects of medium-induced gluon emissions and jet-induced medium excitation on jet observables are systematically examined. Rescatterings of the radiated gluons and recoil partons with the QGP are found essential to account for the enhancement of soft particle yield toward the edge of the jet cone.

nucl-th↗