arXiv subjects

Xuelong Li

Publications and source records attributed to Xuelong Li.

At least 181 records · Page 10Linked to original sources

arXiv2025

Rare Word Recognition and Translation Without Fine-Tuning via Task Vector in Speech Models

Rare words remain a critical bottleneck for speech-to-text systems. While direct fine-tuning improves recognition of target words, it often incurs high cost, catastrophic forgetting, and limited scalability. To address these challenges, we propose a training-free paradigm based on task vectors for rare word recognition and translation. By defining task vectors as parameter differences and introducing word-level task vector arithmetic, our approach enables flexible composition of rare-word capabilities, greatly enhancing scalability and reusability. Extensive experiments across multiple domains show that the proposed method matches or surpasses fine-tuned models on target words, improves general performance by about 5 BLEU, and mitigates catastrophic forgetting.

eess.AS↗

arXiv2025

Beyond Heuristics: A Decision-Theoretic Framework for Agent Memory Management

External memory is a key component of modern large language model (LLM) systems, enabling long-term interaction and personalization. Despite its importance, memory management is still largely driven by hand-designed heuristics, offering little insight into the long-term and uncertain consequences of memory decisions. In practice, choices about what to read or write shape future retrieval and downstream behavior in ways that are difficult to anticipate. We argue that memory management should be viewed as a sequential decision-making problem under uncertainty, where the utility of memory is delayed and dependent on future interactions. To this end, we propose DAM (Decision-theoretic Agent Memory), a decision-theoretic framework that decomposes memory management into immediate information access and hierarchical storage maintenance. Within this architecture, candidate operations are evaluated via value functions and uncertainty estimators, enabling an aggregate policy to arbitrate decisions based on estimated long-term utility and risk. Our contribution is not a new algorithm, but a principled reframing that clarifies the limitations of heuristic approaches and provides a foundation for future research on uncertainty-aware memory systems.

cs.CL↗

arXiv2025

RecTok: Reconstruction Distillation along Rectified Flow

Visual tokenizers play a crucial role in diffusion models. The dimensionality of latent space governs both reconstruction fidelity and the semantic expressiveness of the latent feature. However, a fundamental trade-off is inherent between dimensionality and generation quality, constraining existing methods to low-dimensional latent spaces. Although recent works have leveraged vision foundation models to enrich the semantics of visual tokenizers and accelerate convergence, high-dimensional tokenizers still underperform their low-dimensional counterparts. In this work, we propose RecTok, which overcomes the limitations of high-dimensional visual tokenizers through two key innovations: flow semantic distillation and reconstruction--alignment distillation. Our key insight is to make the forward flow in flow matching semantically rich, which serves as the training space of diffusion transformers, rather than focusing on the latent space as in previous works. Specifically, our method distills the semantic information in VFMs into the forward flow trajectories in flow matching. And we further enhance the semantics by introducing a masked feature reconstruction loss. Our RecTok achieves superior image reconstruction, generation quality, and discriminative performance. It achieves state-of-the-art results on the gFID-50K under both with and without classifier-free guidance settings, while maintaining a semantically rich latent space structure. Furthermore, as the latent dimensionality increases, we observe consistent improvements. Code and model are available at https://shi-qingyu.github.io/rectok.github.io.

cs.CV↗

arXiv2025

FusAD: Time-Frequency Fusion with Adaptive Denoising for General Time Series Analysis

Time series analysis plays a vital role in fields such as finance, healthcare, industry, and meteorology, underpinning key tasks including classification, forecasting, and anomaly detection. Although deep learning models have achieved remarkable progress in these areas in recent years, constructing an efficient, multi-task compatible, and generalizable unified framework for time series analysis remains a significant challenge. Existing approaches are often tailored to single tasks or specific data types, making it difficult to simultaneously handle multi-task modeling and effectively integrate information across diverse time series types. Moreover, real-world data are often affected by noise, complex frequency components, and multi-scale dynamic patterns, which further complicate robust feature extraction and analysis. To ameliorate these challenges, we propose FusAD, a unified analysis framework designed for diverse time series tasks. FusAD features an adaptive time-frequency fusion mechanism, integrating both Fourier and Wavelet transforms to efficiently capture global-local and multi-scale dynamic features. With an adaptive denoising mechanism, FusAD automatically senses and filters various types of noise, highlighting crucial sequence variations and enabling robust feature extraction in complex environments. In addition, the framework integrates a general information fusion and decoding structure, combined with masked pre-training, to promote efficient learning and transfer of multi-granularity representations. Extensive experiments demonstrate that FusAD consistently outperforms state-of-the-art models on mainstream time series benchmarks for classification, forecasting, and anomaly detection tasks, while maintaining high efficiency and scalability. Code is available at https://github.com/zhangda1018/FusAD.

cs.LG↗

arXiv2025

TeleEgo: Benchmarking Egocentric AI Assistants in the Wild

Egocentric AI assistants in real-world settings must process multi-modal inputs (video, audio, text), respond in real time, and retain evolving long-term memory. However, existing benchmarks typically evaluate these abilities in isolation, lack realistic streaming scenarios, or support only short-term tasks. We introduce \textbf{TeleEgo}, a long-duration, streaming, omni-modal benchmark for evaluating egocentric AI assistants in realistic daily contexts. The dataset features over 14 hours per participant of synchronized egocentric video, audio, and text across four domains: work \& study, lifestyle \& routines, social activities, and outings \& culture. All data is aligned on a unified global timeline and includes high-quality visual narrations and speech transcripts, curated through human refinement.TeleEgo defines 12 diagnostic subtasks across three core capabilities: Memory (recalling past events), Understanding (interpreting the current moment), and Cross-Memory Reasoning (linking distant events). It contains 3,291 human-verified QA items spanning multiple question formats (single-choice, binary, multi-choice, and open-ended), evaluated strictly in a streaming setting. We propose Real-Time Accuracy (RTA) to jointly capture correctness and responsiveness under tight decision windows, and Memory Persistence Time (MPT) as a forward-looking metric for long-term retention in continuous streams. In this work, we report RTA results for current models and release TeleEgo, together with an MPT evaluation framework, as a realistic and extensible benchmark for future egocentric assistants with stronger streaming memory, enabling systematic study of both real-time behavior and long-horizon memory.

cs.CV↗

arXiv2025

Aetheria: A multimodal interpretable content safety framework based on multi-agent debate and collaboration

The exponential growth of digital content presents significant challenges for content safety. Current moderation systems, often based on single models or fixed pipelines, exhibit limitations in identifying implicit risks and providing interpretable judgment processes. To address these issues, we propose Aetheria, a multimodal interpretable content safety framework based on multi-agent debate and collaboration.Employing a collaborative architecture of five core agents, Aetheria conducts in-depth analysis and adjudication of multimodal content through a dynamic, mutually persuasive debate mechanism, which is grounded by RAG-based knowledge retrieval.Comprehensive experiments on our proposed benchmark (AIR-Bench) validate that Aetheria not only generates detailed and traceable audit reports but also demonstrates significant advantages over baselines in overall content safety accuracy, especially in the identification of implicit risks. This framework establishes a transparent and interpretable paradigm, significantly advancing the field of trustworthy AI content moderation.

cs.AI↗

arXiv2025

Exploring the Potential of Encoder-free Architectures in 3D LMMs

Encoder-free architectures have been preliminarily explored in the 2D Large Multimodal Models (LMMs), yet it remains an open question whether they can be effectively applied to 3D understanding scenarios. In this paper, we present the first comprehensive investigation into the potential of encoder-free architectures to alleviate the challenges of encoder-based 3D LMMs. These long-standing challenges include the failure to adapt to varying point cloud resolutions during inference and the point features from the encoder not meeting the semantic needs of Large Language Models (LLMs). We identify key aspects for 3D LMMs to remove the pre-trained encoder and enable the LLM to assume the role of the 3D encoder: 1) We propose the LLM-embedded Semantic Encoding strategy in the pre-training stage, exploring the effects of various point cloud self-supervised losses. And we present the Hybrid Semantic Loss to extract high-level semantics. 2) We introduce the Hierarchical Geometry Aggregation strategy in the instruction tuning stage. This incorporates inductive bias into the LLM layers to focus on the local details of the point clouds. To the end, we present the first Encoder-free 3D LMM, ENEL. Our 7B model rivals the state-of-the-art model, PointLLM-PiSA-13B, achieving 57.91%, 61.0%, and 55.20% on the classification, captioning, and VQA tasks, respectively. Our results show that the encoder-free architecture is highly promising for replacing encoder-based architectures in the field of 3D understanding. The code is released at https://github.com/Ivan-Tang-3D/ENEL

cs.CV↗

arXiv2025

TeleAI-Safety: A comprehensive LLM jailbreaking benchmark towards attacks, defenses, and evaluations

While the deployment of large language models (LLMs) in high-value industries continues to expand, the systematic assessment of their safety against jailbreak and prompt-based attacks remains insufficient. Existing safety evaluation benchmarks and frameworks are often limited by an imbalanced integration of core components (attack, defense, and evaluation methods) and an isolation between flexible evaluation frameworks and standardized benchmarking capabilities. These limitations hinder reliable cross-study comparisons and create unnecessary overhead for comprehensive risk assessment. To address these gaps, we present TeleAI-Safety, a modular and reproducible framework coupled with a systematic benchmark for rigorous LLM safety evaluation. Our framework integrates a broad collection of 19 attack methods (including one self-developed method), 29 defense methods, and 19 evaluation methods (including one self-developed method). With a curated attack corpus of 342 samples spanning 12 distinct risk categories, the TeleAI-Safety benchmark conducts extensive evaluations across 14 target models. The results reveal systematic vulnerabilities and model-specific failure cases, highlighting critical trade-offs between safety and utility, and identifying potential defense patterns for future optimization. In practical scenarios, TeleAI-Safety can be flexibly adjusted with customized attack, defense, and evaluation combinations to meet specific demands. We release our complete code and evaluation results to facilitate reproducible research and establish unified safety baselines.

cs.CR↗

arXiv2025

UniDiff: A Unified Diffusion Framework for Multimodal Time Series Forecasting

As multimodal data proliferates across diverse real-world applications, leveraging heterogeneous information such as texts and timestamps for accurate time series forecasting (TSF) has become a critical challenge. While diffusion models demonstrate exceptional performance in generation tasks, their application to TSF remains largely confined to modeling single-modality numerical sequences, overlooking the abundant cross-modal signals inherent in complex heterogeneous data. To address this gap, we propose UniDiff, a unified diffusion framework for multimodal time series forecasting. To process the numerical sequence, our framework first tokenizes the time series into patches, preserving local temporal dynamics by mapping each patch to an embedding space via a lightweight MLP. At its core lies a unified and parallel fusion module, where a single cross-attention mechanism adaptively weighs and integrates structural information from timestamps and semantic context from texts in one step, enabling a flexible and efficient interplay between modalities. Furthermore, we introduce a novel classifier-free guidance mechanism designed for multi-source conditioning, allowing for decoupled control over the guidance strength of textual and temporal information during inference, which significantly enhances model robustness. Extensive experiments on real-world benchmark datasets across eight domains demonstrate that the proposed UniDiff model achieves state-of-the-art performance.

cs.LG↗

arXiv2025

ERF-BA-TFD+: A Multimodal Model for Audio-Visual Deepfake Detection

Deepfake detection is a critical task in identifying manipulated multimedia content. In real-world scenarios, deepfake content can manifest across multiple modalities, including audio and video. To address this challenge, we present ERF-BA-TFD+, a novel multimodal deepfake detection model that combines enhanced receptive field (ERF) and audio-visual fusion. Our model processes both audio and video features simultaneously, leveraging their complementary information to improve detection accuracy and robustness. The key innovation of ERF-BA-TFD+ lies in its ability to model long-range dependencies within the audio-visual input, allowing it to better capture subtle discrepancies between real and fake content. In our experiments, we evaluate ERF-BA-TFD+ on the DDL-AV dataset, which consists of both segmented and full-length video clips. Unlike previous benchmarks, which focused primarily on isolated segments, the DDL-AV dataset allows us to assess the model's performance in a more comprehensive and realistic setting. Our method achieves state-of-the-art results on this dataset, outperforming existing techniques in terms of both accuracy and processing speed. The ERF-BA-TFD+ model demonstrated its effectiveness in the "Workshop on Deepfake Detection, Localization, and Interpretability," Track 2: Audio-Visual Detection and Localization (DDL-AV), and won first place in this competition.

cs.AI↗

arXiv2025

TempoMaster: Efficient Long Video Generation via Next-Frame-Rate Prediction

We present TempoMaster, a novel framework that formulates long video generation as next-frame-rate prediction. Specifically, we first generate a low-frame-rate clip that serves as a coarse blueprint of the entire video sequence, and then progressively increase the frame rate to refine visual details and motion continuity. During generation, TempoMaster employs bidirectional attention within each frame-rate level while performing autoregression across frame rates, thus achieving long-range temporal coherence while enabling efficient and parallel synthesis. Extensive experiments demonstrate that TempoMaster establishes a new state-of-the-art in long video generation, excelling in both visual and temporal quality.

cs.CV↗

arXiv2025

Steering Vision-Language-Action Models as Anti-Exploration: A Test-Time Scaling Approach

Vision-Language-Action (VLA) models, trained via flow-matching or diffusion objectives, excel at learning complex behaviors from large-scale, multi-modal datasets (e.g., human teleoperation, scripted policies). However, since VLAs incorporate diverse data modes in the pre-training stage, and the finetuning dataset often contains demonstration data collected in a kinematically suboptimal or undesirable way, it exists redundant action modes that are irrelevant to the success action modes of the downstream task. Specifically, we observe a critical inference-time fragility among various sampled noises after supervised finetuning of pre-trained VLAs. In this paper, we attribute this instability to the distribution shift between the VLA policy and the policy induced by stable success modes of the downstream task dataset. Thus, we propose \textbf{TACO}, a test-time-scaling (TTS) framework that applies a lightweight pseudo-count estimator as a high-fidelity verifier of action chunks. The VLA models integrated with TACO can execute the actions with maximum pseudo-count from all sampled action chunks, thereby preventing distribution shifts while preserving the generalization ability of VLAs since the constraint is applied only during inference. Our method resembles the classical anti-exploration principle in offline reinforcement learning (RL), and being gradient-free, it incurs significant computational benefits compared to RL update, especially for flow or diffusion-based VLAs which are difficult to perform RL update due to denoising process. Extensive experiments across four simulation benchmarks (RoboTwin2.0, Robotwin, LIBERO, SimplerEnv) and a dual-arm platform demonstrate that our method significantly improves the inference stability and success rates in downstream-task adaptations.

cs.RO↗

arXiv2025

CtrlVDiff: Controllable Video Generation via Unified Multimodal Video Diffusion

We tackle the dual challenges of video understanding and controllable video generation within a unified diffusion framework. Our key insights are two-fold: geometry-only cues (e.g., depth, edges) are insufficient: they specify layout but under-constrain appearance, materials, and illumination, limiting physically meaningful edits such as relighting or material swaps and often causing temporal drift. Enriching the model with additional graphics-based modalities (intrinsics and semantics) provides complementary constraints that both disambiguate understanding and enable precise, predictable control during generation. However, building a single model that uses many heterogeneous cues introduces two core difficulties. Architecturally, the model must accept any subset of modalities, remain robust to missing inputs, and inject control signals without sacrificing temporal consistency. Data-wise, training demands large-scale, temporally aligned supervision that ties real videos to per-pixel multimodal annotations. We then propose CtrlVDiff, a unified diffusion model trained with a Hybrid Modality Control Strategy (HMCS) that routes and fuses features from depth, normals, segmentation, edges, and graphics-based intrinsics (albedo, roughness, metallic), and re-renders videos from any chosen subset with strong temporal coherence. To enable this, we build MMVideo, a hybrid real-and-synthetic dataset aligned across modalities and captions. Across understanding and generation benchmarks, CtrlVDiff delivers superior controllability and fidelity, enabling layer-wise edits (relighting, material adjustment, object insertion) and surpassing state-of-the-art baselines while remaining robust when some modalities are unavailable.

cs.CV↗

arXiv2025

FAIM: Frequency-Aware Interactive Mamba for Time Series Classification

Time series classification (TSC) is crucial in numerous real-world applications, such as environmental monitoring, medical diagnosis, and posture recognition. TSC tasks require models to effectively capture discriminative information for accurate class identification. Although deep learning architectures excel at capturing temporal dependencies, they often suffer from high computational cost, sensitivity to noise perturbations, and susceptibility to overfitting on small-scale datasets. To address these challenges, we propose FAIM, a lightweight Frequency-Aware Interactive Mamba model. Specifically, we introduce an Adaptive Filtering Block (AFB) that leverages Fourier Transform to extract frequency-domain features from time series data. The AFB incorporates learnable adaptive thresholds to dynamically suppress noise and employs element-wise coupling of global and local semantic adaptive filtering, enabling in-depth modeling of the synergy among different frequency components. Furthermore, we design an Interactive Mamba Block (IMB) to facilitate efficient multi-granularity information interaction, balancing the extraction of fine-grained discriminative features and comprehensive global contextual information, thereby endowing FAIM with powerful and expressive representations for TSC tasks. Additionally, we incorporate a self-supervised pre-training mechanism to enhance FAIM's understanding of complex temporal patterns and improve its robustness across various domains and high-noise scenarios. Extensive experiments on multiple benchmarks demonstrate that FAIM consistently outperforms existing state-of-the-art (SOTA) methods, achieving a superior trade-off between accuracy and efficiency and exhibits outstanding performance.

cs.LG↗

arXiv2025

Introducing Visual Scenes and Reasoning: A More Realistic Benchmark for Spoken Language Understanding

Spoken Language Understanding (SLU) consists of two sub-tasks: intent detection (ID) and slot filling (SF). Given its broad range of real-world applications, enhancing SLU for practical deployment is increasingly critical. Profile-based SLU addresses ambiguous user utterances by incorporating context awareness (CA), user profiles (UP), and knowledge graphs (KG) to support disambiguation, thereby advancing SLU research toward real-world applicability. However, existing SLU datasets still fall short in representing real-world scenarios. Specifically, (1) CA uses one-hot vectors for representation, which is overly idealized, and (2) models typically focuses solely on predicting intents and slot labels, neglecting the reasoning process that could enhance performance and interpretability. To overcome these limitations, we introduce VRSLU, a novel SLU dataset that integrates both Visual images and explicit Reasoning. For over-idealized CA, we use GPT-4o and FLUX.1-dev to generate images reflecting users' environments and statuses, followed by human verification to ensure quality. For reasoning, GPT-4o is employed to generate explanations for predicted labels, which are then refined by human annotators to ensure accuracy and coherence. Additionally, we propose an instructional template, LR-Instruct, which first predicts labels and then generates corresponding reasoning. This two-step approach helps mitigate the influence of reasoning bias on label prediction. Experimental results confirm the effectiveness of incorporating visual information and highlight the promise of explicit reasoning in advancing SLU.

cs.AI↗

arXiv2025

UniModel: A Visual-Only Framework for Unified Multimodal Understanding and Generation

We present UniModel, a unified generative model that jointly supports visual understanding and visual generation within a single pixel-to-pixel diffusion framework. Our goal is to achieve unification along three axes: the model, the tasks, and the representations. At the representation level, we eliminate modality discrepancies by mapping both text and images into a shared visual space: textual prompts are rendered as painted text images on a clean canvas, and all inputs and outputs are treated purely as RGB pixels. This yields a fully vision-native formulation of multimodal learning. At the task level, a broad range of vision-language problems are cast as pixel-to-pixel transformations in this visual space. For understanding tasks, the model takes an RGB image and produces a painted text image that visually encodes the semantic prediction. For generation tasks, painted text images serve as visual conditions that guide realistic and semantically aligned image synthesis. Captioning and text-to-image generation thus become different directions of the same underlying visual translation process. At the model level, we instantiate a single Unified Diffusion Transformer trained with rectified flow in pixel space. A shared backbone jointly learns bidirectional mappings between natural images and painted text images, with lightweight task embeddings to specify the desired direction. Experiments on text-to-image synthesis and image-to-text understanding demonstrate strong cross-modal alignment and emergent controllability such as cycle-consistent image-caption-image loops. Our initial exploration suggests that unifying model, tasks, and representations in a single visual space is a promising paradigm for general-purpose multimodal intelligence.

cs.CV↗

arXiv2025

GRPO-RM: Fine-Tuning Representation Models via GRPO-Driven Reinforcement Learning

The Group Relative Policy Optimization (GRPO), a reinforcement learning method used to fine-tune large language models (LLMs), has proved its effectiveness in practical applications such as DeepSeek-R1. It raises a question whether GRPO can be generalized to representation learning models. In this paper, we propose Group Relative Policy Optimization for Representation Model (GRPO-RM), and investigate the performance of GRPO-like policy in post-training representation models. Specifically, our method establishes a predefined output set to functionally replace token sequence sampling in LLMs, thereby generating an output group, which is essential for the probability-driven optimization of GRPO. In addition, a specialized reward function is designed to accommodate the properties of representation models. Extensive experiments are conducted on various real-world datasets to validate the effectiveness of our proposed method.

cs.LG↗

arXiv2025

QSentry: Backdoor Detection for Quantum Neural Networks via Measurement Clustering

Quantum neural networks (QNNs) are an important model for implementing quantum machine learning (QML), while they demonstrate a high degree of vulnerability to backdoor attacks similar to classical networks. To address this issue, a quantum backdoor attack detection framework called QSentry is proposed, in which a quantum Measurement Clustering method is introduced to detect backdoors by identifying statistical anomalies in measurement outputs. It is demonstrated that QSentry can effectively detect anomalous distributions induced by backdoor samples with extensive experiments. It achieves a 75.8% F1 score even under a 1% poisoning rate, and further improves to 85.7% and 93.2% as the poisoning rate increases to 5% and 10%, respectively. The integration of silhouette coefficients and relative cluster size enable QSentry to precisely isolate backdoor samples, yielding estimates that closely match actual poisoning ratios. Evaluations under various quantum attack scenarios demonstrate that QSentry delivers superior robustness and accuracy compared with three state-of-the-art detection methods. This work establishes a practical and effective framework for mitigating backdoor threats in QML.

quant-ph↗