arXiv subjects

Peng Liu

Publications and source records attributed to Peng Liu.

At least 91 records · Page 5Linked to original sources

arXiv2026

FDLLM: A Dedicated Detector for Black-Box LLMs Fingerprinting

Large Language Models (LLMs) are rapidly transforming the landscape of digital content creation. However, the prevalent black-box Application Programming Interface (API) access to many LLMs introduces significant challenges in accountability, governance, and security. LLM fingerprinting, which aims to identify the source model by analyzing statistical and stylistic features of generated text, offers a potential solution. Current progress in this area is hindered by a lack of dedicated datasets and the need for efficient, practical methods that are robust against adversarial manipulations. To address these challenges, we introduce FD-Dataset, a comprehensive bilingual fingerprinting benchmark comprising 90,000 text samples from 20 famous proprietary and open-source LLMs. Furthermore, we present FDLLM, a novel fingerprinting method that leverages parameter-efficient Low-Rank Adaptation (LoRA) to fine-tune a foundation model. This approach enables LoRA to extract deep, persistent features that characterize each source LLM. Through our analysis, we find that LoRA adaptation promotes the aggregation of outputs from the same LLM in representation space while enhancing the separation between different LLMs. This mechanism explains why LoRA proves particularly effective for LLM fingerprinting. Extensive empirical evaluations on FD-Dataset demonstrate FDLLM's superiority, achieving a Macro F1 score 22.1% higher than the strongest baseline. FDLLM also exhibits strong generalization to newly released models, achieving an average accuracy of 95% on unseen models. Notably, FDLLM remains consistently robust under various adversarial attacks, including polishing, translation, and synonym substitution. Experimental results show that FDLLM reduces the average attack success rate from 49.2% (LM-D) to 23.9%.

cs.CR↗

arXiv2026

NorwAI's Large Language Models: Technical Report

Norwegian, spoken by approximately five million people, remains underrepresented in many of the most significant breakthroughs in Natural Language Processing (NLP). To address this gap, the NorLLM team at NorwAI has developed a family of models specifically tailored to Norwegian and other Scandinavian languages, building on diverse Transformer-based architectures such as GPT, Mistral, Llama2, Mixtral and Magistral. These models are either pretrained from scratch or continually pretrained on 25B - 88.45B tokens, using a Norwegian-extended tokenizer and advanced post-training strategies to optimize performance, enhance robustness, and improve adaptability across various real-world tasks. Notably, instruction-tuned variants (e.g., Mistral-7B-Instruct and Mixtral-8x7B-Instruct) showcase strong assistant-style capabilities, underscoring their potential for practical deployment in interactive and domain-specific applications. The NorwAI large language models are openly available to Nordic organizations, companies and students for both research and experimental use. This report provides detailed documentation of the model architectures, training data, tokenizer design, fine-tuning strategies, deployment, and evaluations.

cs.CL↗

arXiv2026

HaluMem: Evaluating Hallucinations in Memory Systems of Agents

Memory systems are key components that enable AI systems such as LLMs and AI agents to achieve long-term learning and sustained interaction. However, during memory storage and retrieval, these systems frequently exhibit memory hallucinations, including fabrication, errors, conflicts, and omissions. Existing evaluations of memory hallucinations are primarily end-to-end question answering, which makes it difficult to localize the operational stage within the memory system where hallucinations arise. To address this, we introduce the Hallucination in Memory Benchmark (HaluMem), the first operation level hallucination evaluation benchmark tailored to memory systems. HaluMem defines three evaluation tasks (memory extraction, memory updating, and memory question answering) to comprehensively reveal hallucination behaviors across different operational stages of interaction. To support evaluation, we construct user-centric, multi-turn human-AI interaction datasets, HaluMem-Medium and HaluMem-Long. Both include about 15k memory points and 3.5k multi-type questions. The average dialogue length per user reaches 1.5k and 2.6k turns, with context lengths exceeding 1M tokens, enabling evaluation of hallucinations across different context scales and task complexities. Empirical studies based on HaluMem show that existing memory systems tend to generate and accumulate hallucinations during the extraction and updating stages, which subsequently propagate errors to the question answering stage. Future research should focus on developing interpretable and constrained memory operation mechanisms that systematically suppress hallucinations and improve memory reliability.

cs.CL↗

arXiv2026

NextFlow: Unified Sequential Modeling Activates Multimodal Understanding and Generation

We present NextFlow, a unified decoder-only autoregressive transformer trained on 6 trillion interleaved text-image discrete tokens. By leveraging a unified vision representation within a unified autoregressive architecture, NextFlow natively activates multimodal understanding and generation capabilities, unlocking abilities of image editing, interleaved content and video generation. Motivated by the distinct nature of modalities - where text is strictly sequential and images are inherently hierarchical - we retain next-token prediction for text but adopt next-scale prediction for visual generation. This departs from traditional raster-scan methods, enabling the generation of 1024x1024 images in just 5 seconds - orders of magnitude faster than comparable AR models. We address the instabilities of multi-scale generation through a robust training recipe. Furthermore, we introduce a prefix-tuning strategy for reinforcement learning. Experiments demonstrate that NextFlow achieves state-of-the-art performance among unified models and rivals specialized diffusion baselines in visual quality.

cs.CV↗

arXiv2025

Step-DeepResearch Technical Report

As LLMs shift toward autonomous agents, Deep Research has emerged as a pivotal metric. However, existing academic benchmarks like BrowseComp often fail to meet real-world demands for open-ended research, which requires robust skills in intent recognition, long-horizon decision-making, and cross-source verification. To address this, we introduce Step-DeepResearch, a cost-effective, end-to-end agent. We propose a Data Synthesis Strategy Based on Atomic Capabilities to reinforce planning and report writing, combined with a progressive training path from agentic mid-training to SFT and RL. Enhanced by a Checklist-style Judger, this approach significantly improves robustness. Furthermore, to bridge the evaluation gap in the Chinese domain, we establish ADR-Bench for realistic deep research scenarios. Experimental results show that Step-DeepResearch (32B) scores 61.4% on Scale AI Research Rubrics. On ADR-Bench, it significantly outperforms comparable models and rivals SOTA closed-source models like OpenAI and Gemini DeepResearch. These findings prove that refined training enables medium-sized models to achieve expert-level capabilities at industry-leading cost-efficiency.

cs.CL↗

arXiv2025

DiTSinger: Scaling Singing Voice Synthesis with Diffusion Transformer and Implicit Alignment

Recent progress in diffusion-based Singing Voice Synthesis (SVS) demonstrates strong expressiveness but remains limited by data scarcity and model scalability. We introduce a two-stage pipeline: a compact seed set of human-sung recordings is constructed by pairing fixed melodies with diverse LLM-generated lyrics, and melody-specific models are trained to synthesize over 500 hours of high-quality Chinese singing data. Building on this corpus, we propose DiTSinger, a Diffusion Transformer with RoPE and qk-norm, systematically scaled in depth, width, and resolution for enhanced fidelity. Furthermore, we design an implicit alignment mechanism that obviates phoneme-level duration labels by constraining phoneme-to-acoustic attention within character-level spans, thereby improving robustness under noisy or uncertain alignments. Extensive experiments validate that our approach enables scalable, alignment-free, and high-fidelity SVS.

cs.SD↗

arXiv2025

UDFS: Lightweight Representation-Driven Open World Robust Encrypted Traffic Classification

In recent years, sequence features such as packet length have received considerable attention due to their central role in encrypted traffic analysis. Existing sequence modeling approaches can be broadly categorized into flow-level and trace-level methods: the former suffer from high feature redundancy, limiting their discriminative power, whereas the latter preserve complete information but incur substantial computational and storage overhead. To address these limitations, we propose the \textbf{U}p-\textbf{D}own \textbf{F}low \textbf{S}equence (\textbf{UDFS}) representation, which compresses an entire trace into a two-dimensional sequence and characterizes each flow by the aggregate of its upstream and downstream traffic, reducing complexity while maintaining high discriminability. Furthermore, to address the challenge of class-specific discriminability differences, we propose an adaptive threshold mechanism that dynamically adjusts training weights and rejection boundaries, enhancing the model's classification performance. Experimental results demonstrate that the proposed method achieves superior classification performance and robustness on both coarse-grained and fine-grained datasets, as well as under concept drift and open-world scenarios. Code and Dataset are available at https://github.com/kid1999/UDFS.

cs.NI↗

arXiv2025

Pareto-optimal reinsurance under dependence uncertainty

This paper studies Pareto-optimal reinsurance design in a monopolistic market with multiple primary insurers and a single reinsurer, all with heterogeneous risk preferences. The risk preferences are characterized by a family of risk measures, called Range Value-at-Risk (RVaR), which includes both Value-at-Risk (VaR) and Expected Shortfall (ES) as special cases. Recognizing the practical difficulty of accurately estimating the dependence structure among the insurers' losses, we adopt a robust optimization approach that assumes the marginal distributions are known while leaving the dependence structure unspecified. We provide a complete characterization of optimal indemnity schedules under the worst-case scenario, showing that the infinite-dimensional optimization problem can be reduced to a tractable finite-dimensional problem involving only two or three parameters for each indemnity function. Additionally, for independent and identically distributed risks, we exploit the argument of asymptotic normality to derive optimal two-parameter layer contracts. Finally, numerical applications are considered in a two-insurer setting to illustrate the influence of the dependence structures and heterogeneous risk tolerances on optimal strategies and the corresponding risk evaluation.

q-fin.RM↗

arXiv2025

VFM-ISRefiner: Towards Better Adapting Vision Foundation Models for Interactive Segmentation of Remote Sensing Images

Interactive image segmentation(IIS) plays a critical role in generating precise annotations for remote sensing imagery, where objects often exhibit scale variations, irregular boundaries and complex backgrounds. However, existing IIS methods, primarily designed for natural images, struggle to generalize to remote sensing domains due to limited annotated data and computational overhead. To address these challenges, we proposed RS-ISRefiner, a novel click-based IIS framework tailored for remote sensing images. The framework employs an adapter-based tuning strategy that preserves the general representations of Vision Foundation Models while enabling efficient learning of remote sensing-specific spatial and boundary characteristics. A hybrid attention mechanism integrating convolutional local modeling with Transformer-based global reasoning enhances robustness against scale diversity and scene complexity. Furthermore, an improved probability map modulation scheme effectively incorporates historical user interactions, yielding more stable iterative refinement and higher boundary accuracy. Comprehensive experiments on six remote sensing datasets, including iSAID, ISPRS Potsdam, SandBar, NWPU, LoveDA Urban and WHUBuilding, demonstrate that RS-ISRefiner consistently outperforms state-of-the-art IIS methods in terms of segmentation accuracy, efficiency and interaction cost. These results confirm the effectiveness and generalizability of our framework, making it highly suitable for high-quality instance segmentation in practical remote sensing scenarios. The codes are available at https://github.com/wondelyan/VFM-ISRefiner .

cs.CV↗

arXiv2025

Towards Data-efficient Customer Intent Recognition with Prompt-based Learning Paradigm

Recognizing customer intent accurately with language models based on customer-agent conversational data is essential in today's digital customer service marketplace, but it is often hindered by the lack of sufficient labeled data. In this paper, we introduce the prompt-based learning paradigm that significantly reduces the dependency on extensive datasets. Utilizing prompted training combined with answer mapping techniques, this approach allows small language models to achieve competitive intent recognition performance with only a minimal amount of training data. Furthermore, We enhance the performance by integrating active sampling and ensemble learning strategies in the prompted training pipeline. Additionally, preliminary tests in a zero-shot setting demonstrate that, with well-crafted and detailed prompts, small language models show considerable instruction-following potential even without any further training. These results highlight the viability of semantic modeling of conversational data in a more data-efficient manner with minimal data use, paving the way for advancements in AI-driven customer service.

cs.CL↗

arXiv2025

EnhancedRL: An Enhanced-State Reinforcement Learning Algorithm for Multi-Task Fusion in Recommender Systems

As a key stage of Recommender Systems (RSs), Multi-Task Fusion (MTF) is responsible for merging multiple scores output by Multi-Task Learning (MTL) into a single score, finally determining the recommendation results. Recently, Reinforcement Learning (RL) has been applied to MTF to maximize long-term user satisfaction within a recommendation session. However, due to limitations in modeling paradigm, all existing RL algorithms for MTF can only utilize user features and statistical features as the state to generate actions at the user level, but unable to leverage item features and other valuable features, which leads to suboptimal performance. Overcoming this problem requires a breakthrough in the existing modeling paradigm, yet, to date, no prior work has addressed it. To tackle this challenge, we propose EnhancedRL, an innovative RL algorithm. Unlike existing RL-MTF methods, EnhancedRL takes the enhanced state as input, incorporating not only user features but also item features and other valuable information. Furthermore, it introduces a tailored actor-critic framework - including redesigned actor and critics and a novel learning procedure - to optimize long-term rewards at the user-item pair level within a recommendation session. Extensive offline and online experiments are conducted in an industrial RS and the results demonstrate that EnhancedRL outperforms other methods remarkably, achieving a +3.84% increase in user valid consumption and a +0.58% increase in user duration time. To the best of our knowledge, EnhancedRL is the first work to address this challenge, and it has been fully deployed in a large-scale RS since September 14, 2023, yielding significant improvements.

cs.IR↗

arXiv2025

Internal superfluid response and torque evolution in the giant glitch of PSR J1718-3718

We investigate the post-glitch rotational evolution of pulsars by analyzing the 2007 giant glitch of PSR J1718$-$3718 using a vortex creep model that incorporates both inward and outward nonlinear vortex motion, along with a time-varying external torque. A comprehensive fitting framework is developed, constrained by prior knowledge of moment of inertia participation from previous glitch studies. We apply a Markov Chain Monte Carlo approach to quantify uncertainties and parameter correlations. The model reproduces the observed timing data and yields physically consistent values for moment of inertia fractions and creep timescales. Our results indicate that inward creep and a long-term change in external torque dominate the observed increase in spin-down rate, pointing to structural changes within the star-likely triggered by a crustquake that initiated both vortex motion and a change in the moment of inertia. We estimate that the glitch involved approximately $2.4 \times 10^{12}$ inward-moving vortices and $\sim 142$ crustal plates with a typical size of $\sim 0.03$ km. This study demonstrates that detailed post-glitch modeling of sparse timing data can simultaneously constrain internal superfluid dynamics and external torque evolution, providing a quantitative framework to probe the structural properties of neutron star interiors.

astro-ph.HE↗

arXiv2025

MemOS: A Memory OS for AI System

Large Language Models (LLMs) have become an essential infrastructure for Artificial General Intelligence (AGI), yet their lack of well-defined memory management systems hinders the development of long-context reasoning, continual personalization, and knowledge consistency.Existing models mainly rely on static parameters and short-lived contextual states, limiting their ability to track user preferences or update knowledge over extended periods.While Retrieval-Augmented Generation (RAG) introduces external knowledge in plain text, it remains a stateless workaround without lifecycle control or integration with persistent representations.Recent work has modeled the training and inference cost of LLMs from a memory hierarchy perspective, showing that introducing an explicit memory layer between parameter memory and external retrieval can substantially reduce these costs by externalizing specific knowledge. Beyond computational efficiency, LLMs face broader challenges arising from how information is distributed over time and context, requiring systems capable of managing heterogeneous knowledge spanning different temporal scales and sources. To address this challenge, we propose MemOS, a memory operating system that treats memory as a manageable system resource. It unifies the representation, scheduling, and evolution of plaintext, activation-based, and parameter-level memories, enabling cost-efficient storage and retrieval. As the basic unit, a MemCube encapsulates both memory content and metadata such as provenance and versioning. MemCubes can be composed, migrated, and fused over time, enabling flexible transitions between memory types and bridging retrieval with parameter-based learning. MemOS establishes a memory-centric system framework that brings controllability, plasticity, and evolvability to LLMs, laying the foundation for continual learning and personalized modeling.

cs.CL↗

arXiv2025

Leveraging Large-Scale Pretrained Spatial-Spectral Priors for General Zero-Shot Pansharpening

Existing deep learning methods for remote sensing image fusion often suffer from poor generalization when applied to unseen datasets due to the limited availability of real training data and the domain gap between different satellite sensors. To address this challenge, we explore the potential of foundation models by proposing a novel pretraining strategy that leverages large-scale simulated datasets to learn robust spatial-spectral priors. Specifically, our approach first constructs diverse simulated datasets by applying various degradation operations (blur, noise, downsampling) and augmentations (bands generation, channel shuffling, high-pass filtering, color jittering, etc.) to natural images from ImageNet and remote sensing images from SkyScript. We then pretrain fusion models on these simulated data to learn generalizable spatial-spectral representations. The pretrained models are subsequently evaluated on six datasets (WorldView-2/3/4, IKONOS, QuickBird, GaoFen-2) using zero-shot and one-shot paradigms, with both full- and freeze-tuning approaches for fine-tuning. Extensive experiments on different network architectures including convolutional neural networks, Transformer, and Mamba demonstrate that our pretraining strategy significantly improves generalization performance across different satellite sensors and imaging conditions for various fusion models. The pretrained models achieve superior results in zero-shot scenarios and show remarkable adaptation capability with minimal real data in one-shot settings. Our work provides a practical solution for cross-domain pansharpening, establishes a new benchmark for generalization in remote sensing image fusion tasks, and paves the way for leveraging foundation models through advanced training strategies.

cs.CV↗

arXiv2025

Depth Matching Method Based on ShapeDTW for Oil-Based Mud Imager

In well logging operations using the oil-based mud (OBM) microresistivity imager, which employs an interleaved design with upper and lower pad sets, depth misalignment issues persist between the pad images even after velocity correction. This paper presents a depth matching method for borehole images based on the Shape Dynamic Time Warping (ShapeDTW) algorithm. The method extracts local shape features to construct a morphologically sensitive distance matrix, better preserving structural similarity between sequences during alignment. We implement this by employing a combined feature set of the one-dimensional Histogram of Oriented Gradients (HOG1D) and the original signal as the shape descriptor. Field test examples demonstrate that our method achieves precise alignment for images with complex textures, depth shifts, or local scaling. Furthermore, it provides a flexible framework for feature extension, allowing the integration of other descriptors tailored to specific geological features.

cs.CV↗

arXiv2025

Learning Multi-Access Point Coordination in Agentic AI Wi-Fi with Large Language Models

Multi-access point coordination (MAPC) is a key technology for enhancing throughput in next-generation Wi-Fi within dense overlapping basic service sets. However, existing MAPC protocols rely on static, protocol-defined rules, which limits their ability to adapt to dynamic network conditions such as varying interference levels and topologies. To address this limitation, we propose a novel Agentic AI Wi-Fi framework where each access point, modeled as an autonomous large language model agent, collaboratively reasons about the network state and negotiates adaptive coordination strategies in real time. This dynamic collaboration is achieved through a cognitive workflow that enables the agents to engage in natural language dialogue, leveraging integrated memory, reflection, and tool use to ground their decisions in past experience and environmental feedback. Comprehensive simulation results demonstrate that our agentic framework successfully learns to adapt to diverse and dynamic network environments, significantly outperforming the state-of-the-art spatial reuse baseline and validating its potential as a robust and intelligent solution for future wireless networks.

cs.AI↗

arXiv2025

AutoMAS: A Generic Multi-Agent System for Algorithm Self-Adaptation in Wireless Networks

The wireless communication environment has the characteristic of strong dynamics. Conventional wireless networks operate based on the static rules with predefined algorithms, lacking the self-adaptation ability. The rapid development of artificial intelligence (AI) provides a possibility for wireless networks to become more intelligent and fully automated. As such, we plan to integrate the cognitive capability and high intelligence of the emerging AI agents into wireless networks. In this work, we propose AutoMAS, a generic multi-agent system which can autonomously select the most suitable wireless optimization algorithm according to the dynamic wireless environment. Our AutoMAS combines theoretically guaranteed wireless algorithms with agents' perception ability, thereby providing sounder solutions to complex tasks no matter how the environment changes. As an example, we conduct a case study on the classical channel estimation problem, where the mobile user moves in diverse environments with different channel propagation characteristics. Simulation results demonstrate that our AutoMAS can guarantee the highest accuracy in changing scenarios. Similarly, our AutoMAS can be generalized to autonomously handle various tasks in 6G wireless networks with high accuracy.

eess.SP↗

arXiv2025

CST Anti-UAV: A Thermal Infrared Benchmark for Tiny UAV Tracking in Complex Scenes

The widespread application of Unmanned Aerial Vehicles (UAVs) has raised serious public safety and privacy concerns, making UAV perception crucial for anti-UAV tasks. However, existing UAV tracking datasets predominantly feature conspicuous objects and lack diversity in scene complexity and attribute representation, limiting their applicability to real-world scenarios. To overcome these limitations, we present the CST Anti-UAV, a new thermal infrared dataset specifically designed for Single Object Tracking (SOT) in Complex Scenes with Tiny UAVs (CST). It contains 220 video sequences with over 240k high-quality bounding box annotations, highlighting two key properties: a significant number of tiny-sized UAV targets and the diverse and complex scenes. To the best of our knowledge, CST Anti-UAV is the first dataset to incorporate complete manual frame-level attribute annotations, enabling precise evaluations under varied challenges. To conduct an in-depth performance analysis for CST Anti-UAV, we evaluate 20 existing SOT methods on the proposed dataset. Experimental results demonstrate that tracking tiny UAVs in complex environments remains a challenge, as the state-of-the-art method achieves only 35.92% state accuracy, much lower than the 67.69% observed on the Anti-UAV410 dataset. These findings underscore the limitations of existing benchmarks and the need for further advancements in UAV tracking research. The CST Anti-UAV benchmark is about to be publicly released, which not only fosters the development of more robust SOT methods but also drives innovation in anti-UAV systems.

cs.CV↗