arXiv subjects

Yongfeng Huang

Publications and source records attributed to Yongfeng Huang.

At least 37 records · Page 2Linked to original sources

arXiv2025

Long-context Language Models Fail in Basic Retrieval Tasks Without Sufficient Reasoning Steps

Long-context language models (LCLMs), characterized by their extensive context window, are becoming popular. However, despite the fact that they are nearly perfect at standard long-context retrieval tasks, our evaluations demonstrate they fail in some basic cases. Later, we find they can be well addressed with a sufficient number of reasoning steps, guided by specific CoT prompts. This result emphasizes the potential necessity of solving specific long-context tasks using long-CoT methods, while previous long-context benchmarks always ignore the necessity of long reasoning for long-context tasks and treat them as direct QA tasks.

cs.CL↗

arXiv2025

Phase transitions driven by solute concentration, temperature, and pressure in uranium-6wt % niobium alloy

An angular-dependent potential for the U-Nb system is developed based on an existing ADP for U and a new EAM potential for Nb through fitting flexible cross-interaction functions and alloy parameters to experimental and first-principles data, enabling accurate prediction of phase transitions (alpha to gamma driven by solute concentration; alpha-prime to gamma under temperature in U-6Nb alloy), elastic properties, defect energetics, and mixed enthalpy. The potential reliably reproduces melting points of U-Nb solid solutions and captures lattice parameter expansion of gamma U-6Nb. Notably, it correctly predicts Hugoniot relations and equations of state up to about 90 GPa and resolves the alpha-prime to gamma transition under static high pressure. Combined with atomic simulations, we reveal a twinning-coupled alpha-prime to gamma transition of U-6Nb under high pressures: {112}gamma twins form via nanosecond-scale twinning precursors generated during the transient adiabatic compressions. The static phase transition pressure is predicted to be 54.5 GPa, comparable to 67.2 GPa by first-principles calculations. Besides, our result suggests that U-6Nb single-crystal would experience a nonlinear elastic relaxation before yielding plastically at 3.1 GPa (shear stress: 0.9 GPa). The results in this work help resolve long-standing discrepancies in understanding the abnormal shear stress relaxation mechanisms under high-pressure shock loading.

cond-mat.mtrl-sci↗

arXiv2025

Guided Navigation in Knowledge-Dense Environments: Structured Semantic Exploration with Guidance Graphs

While Large Language Models (LLMs) exhibit strong linguistic capabilities, their reliance on static knowledge and opaque reasoning processes limits their performance in knowledge intensive tasks. Knowledge graphs (KGs) offer a promising solution, but current exploration methods face a fundamental trade off: question guided approaches incur redundant exploration due to granularity mismatches, while clue guided methods fail to effectively leverage contextual information for complex scenarios. To address these limitations, we propose Guidance Graph guided Knowledge Exploration (GG Explore), a novel framework that introduces an intermediate Guidance Graph to bridge unstructured queries and structured knowledge retrieval. The Guidance Graph defines the retrieval space by abstracting the target knowledge' s structure while preserving broader semantic context, enabling precise and efficient exploration. Building upon the Guidance Graph, we develop: (1) Structural Alignment that filters incompatible candidates without LLM overhead, and (2) Context Aware Pruning that enforces semantic consistency with graph constraints. Extensive experiments show our method achieves superior efficiency and outperforms SOTA, especially on complex tasks, while maintaining strong performance with smaller LLMs, demonstrating practical value.

cs.CL↗

arXiv2025

Fine-grained Stateful Knowledge Exploration: Effective and Efficient Graph Retrieval with Large Language Models

Large Language Models (LLMs) have shown impressive capabilities, yet updating their knowledge remains a significant challenge, often leading to outdated or inaccurate responses. A proposed solution is the integration of external knowledge bases, such as knowledge graphs, with LLMs. Most existing methods use a paradigm that treats the whole question as the objective, with relevant knowledge being incrementally retrieved from the knowledge graph. However, this paradigm often leads to a granularity mismatch between the target question and the retrieved entities and relations. As a result, the information in the question cannot precisely correspond to the retrieved knowledge. This may cause redundant exploration or omission of vital knowledge, thereby leading to enhanced computational consumption and reduced retrieval accuracy. To address the limitations of coarse-grained knowledge exploration, we propose FiSKE, a novel paradigm for Fine-grained Stateful Knowledge Exploration. FiSKE first decomposes questions into fine-grained clues, then employs an adaptive mapping strategy during knowledge exploration process to resolve ambiguity in clue-to-graph mappings. This strategy dynamically infers contextual correspondences while maintaining a stateful record of the mappings. A clue-driven termination mechanism ensures rigorous augmentation--leveraging fully mapped paths for LLMs while reverting to chain-of-thought reasoning when necessary. Our approach balances precision and efficiency. Experiments on multiple datasets revealed that our paradigm surpasses current advanced methods in knowledge retrieval while significantly reducing the average number of LLM invocations.

cs.CL↗

arXiv2025

Multi-year Polarimetric Monitoring of Four CHIME-Discovered Repeating Fast Radio Bursts with FAST

In this study, we report multi-year polarization measurements of four repeating FRBs initially discovered by CHIME: FRBs~20190117A, 20190208A, 20190303A, and 20190417A. We observed the four repeating FRBs with FAST, detecting a total of 66 bursts. Two bursts from FRB~20190417A exhibit a circular polarization signal-to-noise ratio greater than 7, with the highest circular polarization fraction recorded at 35.7%. While the bursts from FRBs 20190208A and 20190303A are highly linearly polarized, those from FRBs~20190117A and 20190417A show depolarization due to multi-path propagation, with σ_{\mathrm{RM}} = 2.78 \pm 0.05 rad m$^{-2}$ and 5.19 \pm 0.09 rad m$^{-2}$, respectively. The linear polarization distributions among five repeating FRB--FRBs~20190208A, 20190303A, 20201124A, 20220912A, and 20240114A--are nearly identical but show distinct differences from those of non-repeating FRBs. FRBs~20190117A, 20190303A, and 20190417A exhibit substantial rotation measure (RM) variations between bursts, joining other repeating FRBs in this behavior. Combining these findings with published results, 64% of repeating FRBs show RM variations greater than 50 rad m$^{-2}$, and 21\% exhibit RM reversals. A significant proportion of repeating FRBs reside in a dynamic magneto-ionic environment. The structure function of RM variations shows a power-law index of $γ\sim (0-0.8)$, corresponding to a shallow power spectrum $α= -(γ+ 2) \sim -(2.0-2.8)$ of turbulence, if the RM variations are attributed to turbulence. This suggests that the variations are dominated by small-scale RM density fluctuations. We perform K-S tests comparing the RMs of repeating and non-repeating FRBs, which reveal a marginal dichotomy in the distribution of their RMs.We caution that the observed dichotomy may be due to the small sample size and selection biases.

astro-ph.HE↗

arXiv2025

MrM: Black-Box Membership Inference Attacks against Multimodal RAG Systems

Multimodal retrieval-augmented generation (RAG) systems enhance large vision-language models by integrating cross-modal knowledge, enabling their increasing adoption across real-world multimodal tasks. These knowledge databases may contain sensitive information that requires privacy protection. However, multimodal RAG systems inherently grant external users indirect access to such data, making them potentially vulnerable to privacy attacks, particularly membership inference attacks (MIAs). % Existing MIA methods targeting RAG systems predominantly focus on the textual modality, while the visual modality remains relatively underexplored. To bridge this gap, we propose MrM, the first black-box MIA framework targeted at multimodal RAG systems. It utilizes a multi-object data perturbation framework constrained by counterfactual attacks, which can concurrently induce the RAG systems to retrieve the target data and generate information that leaks the membership information. Our method first employs an object-aware data perturbation method to constrain the perturbation to key semantics and ensure successful retrieval. Building on this, we design a counterfact-informed mask selection strategy to prioritize the most informative masked regions, aiming to eliminate the interference of model self-knowledge and amplify attack efficacy. Finally, we perform statistical membership inference by modeling query trials to extract features that reflect the reconstruction of masked semantics from response patterns. Experiments on two visual datasets and eight mainstream commercial visual-language models (e.g., GPT-4o, Gemini-2) demonstrate that MrM achieves consistently strong performance across both sample-level and set-level evaluations, and remains robust under adaptive defenses.

cs.CV↗

arXiv2025

Enhancing Watermarking Quality for LLMs via Contextual Generation States Awareness

Recent advancements in watermarking techniques have enabled the embedding of secret messages into AI-generated text (AIGT), serving as an important mechanism for AIGT detection. Existing methods typically interfere with the generation processes of large language models (LLMs) to embed signals within the generated text. However, these methods often rely on heuristic rules, which can result in suboptimal token selection and a subsequent decline in the quality of the generated content. In this paper, we introduce a plug-and-play contextual generation states-aware watermarking framework (CAW) that dynamically adjusts the embedding process. It can be seamlessly integrated with various existing watermarking methods to enhance generation quality. First, CAW incorporates a watermarking capacity evaluator, which can assess the impact of embedding messages at different token positions by analyzing the contextual generation states. Furthermore, we introduce a multi-branch pre-generation mechanism to avoid the latency caused by the proposed watermarking strategy. Building on this, CAW can dynamically adjust the watermarking process based on the evaluated watermark capacity of each token, thereby minimizing potential degradation in content quality. Extensive experiments conducted on datasets across multiple domains have verified the effectiveness of our method, demonstrating superior performance compared to various baselines in terms of both detection rate and generation quality.

cs.CR↗

arXiv2025

Efficient Diffusion Models: A Survey

Diffusion models have emerged as powerful generative models capable of producing high-quality contents such as images, videos, and audio, demonstrating their potential to revolutionize digital content creation. However, these capabilities come at the cost of their significant computational resources and lengthy generation time, underscoring the critical need to develop efficient techniques for practical deployment. In this survey, we provide a systematic and comprehensive review of research on efficient diffusion models. We organize the literature in a taxonomy consisting of three main categories, covering distinct yet interconnected efficient diffusion model topics from algorithm-level, system-level, and framework perspective, respectively. We have also created a GitHub repository where we organize the papers featured in this survey at https://github.com/AIoT-MLSys-Lab/Efficient-Diffusion-Model-Survey. We hope our survey can serve as a valuable resource to help researchers and practitioners gain a systematic understanding of efficient diffusion model research and inspire them to contribute to this important and exciting field.

cs.LG↗

arXiv2025

TaxAgent: How Large Language Model Designs Fiscal Policy

Economic inequality is a global challenge, intensifying disparities in education, healthcare, and social stability. Traditional systems like the U.S. federal income tax reduce inequality but lack adaptability. Although models like the Saez Optimal Taxation adjust dynamically, they fail to address taxpayer heterogeneity and irrational behavior. This study introduces TaxAgent, a novel integration of large language models (LLMs) with agent-based modeling (ABM) to design adaptive tax policies. In our macroeconomic simulation, heterogeneous H-Agents (households) simulate real-world taxpayer behaviors while the TaxAgent (government) utilizes LLMs to iteratively optimize tax rates, balancing equity and productivity. Benchmarked against Saez Optimal Taxation, U.S. federal income taxes, and free markets, TaxAgent achieves superior equity-efficiency trade-offs. This research offers a novel taxation solution and a scalable, data-driven framework for fiscal policy evaluation.

cs.AI↗

arXiv2025

Mitigate Position Bias in Large Language Models via Scaling a Single Dimension

Large Language Models (LLMs) are increasingly applied in various real-world scenarios due to their excellent generalization capabilities and robust generative abilities. However, they exhibit position bias, also known as "lost in the middle", a phenomenon that is especially pronounced in long-context scenarios, which indicates the placement of the key information in different positions of a prompt can significantly affect accuracy. This paper first explores the micro-level manifestations of position bias, concluding that attention weights are a micro-level expression of position bias. It further identifies that, in addition to position embeddings, causal attention mask also contributes to position bias by creating position-specific hidden states. Based on these insights, we propose a method to mitigate position bias by scaling this positional hidden states. Experiments on the NaturalQuestions Multi-document QA, KV retrieval, LongBench and timeline reorder tasks, using various models including RoPE models, context windowextended models, and Alibi models, demonstrate the effectiveness and generalizability of our approach. Our method can improve performance by up to 15.2% by modifying just one dimension of hidden states. Our code is available at https://aka.ms/PositionalHidden.

cs.CL↗

arXiv2025

Tracking the jet-like corona of black hole Swift J1727.8-1613 during a flare state through Type-C quasi-periodic oscillations

Type-C quasi-periodic oscillations (QPOs) in black hole X-ray transients typically manifest in the low-hard and hard-intermediate states. This study presents a detailed spectral and temporal analysis of the black hole candidate Swift J1727.8-1613 using NICER observations from August and September 2023, with a focus on the first flare period. We detected Type-C QPOs whose centroid frequency increased from 0.33 Hz to 2.63 Hz. An additional increase in frequency was observed when the outburst entered a flare period. The time-averaged spectra, along with the rms and phase-lag spectra of the QPOs, were jointly fitted using the time-dependent Comptonization model vkompthdk to examine the geometry of the corona during this flare. Correlations between spectral and temporal properties suggest that the detected type-C QPOs are primarily modulated by Lense-Thirring precession. Leveraging simultaneous radio observations that indicate discrete jet ejections, we proposed a scenario to describe the co-evolution of the disk-corona-jet during a flare (~3 days). This scenario is partially supported for the first time by polarization data in the soft gamma-ray band from INTEGRAL/IBIS. A phenomenological analysis of the corona scenario was also conducted.

astro-ph.HE↗

arXiv2025

HeteRAG: A Heterogeneous Retrieval-augmented Generation Framework with Decoupled Knowledge Representations

Retrieval-augmented generation (RAG) methods can enhance the performance of LLMs by incorporating retrieved knowledge chunks into the generation process. In general, the retrieval and generation steps usually have different requirements for these knowledge chunks. The retrieval step benefits from comprehensive information to improve retrieval accuracy, whereas excessively long chunks may introduce redundant contextual information, thereby diminishing both the effectiveness and efficiency of the generation process. However, existing RAG methods typically employ identical representations of knowledge chunks for both retrieval and generation, resulting in suboptimal performance. In this paper, we propose a heterogeneous RAG framework (\myname) that decouples the representations of knowledge chunks for retrieval and generation, thereby enhancing the LLMs in both effectiveness and efficiency. Specifically, we utilize short chunks to represent knowledge to adapt the generation step and utilize the corresponding chunk with its contextual information from multi-granular views to enhance retrieval accuracy. We further introduce an adaptive prompt tuning method for the retrieval model to adapt the heterogeneous retrieval augmented generation process. Extensive experiments demonstrate that \myname achieves significant improvements compared to baselines.

cs.IR↗

arXiv2025

pFedGPA: Diffusion-based Generative Parameter Aggregation for Personalized Federated Learning

Federated Learning (FL) offers a decentralized approach to model training, where data remains local and only model parameters are shared between the clients and the central server. Traditional methods, such as Federated Averaging (FedAvg), linearly aggregate these parameters which are usually trained on heterogeneous data distributions, potentially overlooking the complex, high-dimensional nature of the parameter space. This can result in degraded performance of the aggregated model. While personalized FL approaches can mitigate the heterogeneous data issue to some extent, the limitation of linear aggregation remains unresolved. To alleviate this issue, we investigate the generative approach of diffusion model and propose a novel generative parameter aggregation framework for personalized FL, \texttt{pFedGPA}. In this framework, we deploy a diffusion model on the server to integrate the diverse parameter distributions and propose a parameter inversion method to efficiently generate a set of personalized parameters for each client. This inversion method transforms the uploaded parameters into a latent code, which is then aggregated through denoising sampling to produce the final personalized parameters. By encoding the dependence of a client's model parameters on the specific data distribution using the high-capacity diffusion model, \texttt{pFedGPA} can effectively decouple the complexity of the overall distribution of all clients' model parameters from the complexity of each individual client's parameter distribution. Our experimental results consistently demonstrate the superior performance of the proposed method across multiple datasets, surpassing baseline approaches.

cs.LG↗

arXiv2025

Science objectives of the Einstein Probe mission

The Einstein Probe (EP) is an interdisciplinary mission of time-domain and X-ray astronomy. Equipped with a wide-field lobster-eye X-ray focusing imager, EP will discover cosmic X-ray transients and monitor the X-ray variability of known sources in 0.5-4 keV, at a combination of detecting sensitivity and cadence that is not accessible to the previous and current wide-field monitoring missions. EP can perform quick characterisation of transients or outbursts with a Wolter-I X-ray telescope onboard. In this paper, the science objectives of the Einstein Probe mission are presented. EP is expected to enlarge the sample of previously known or predicted but rare types of transients with a wide range of timescales. Among them, fast extragalactic transients will be surveyed systematically in soft X-rays, which include γ-ray bursts and their variants, supernova shock breakouts, and the predicted X-ray transients associated with binary neutron star mergers. EP will detect X-ray tidal disruption events and outbursts from active galactic nuclei, possibly at an early phase of the flares for some. EP will monitor the variability and outbursts of X-rays from white dwarfs, neutron stars and black holes in our and neighbouring galaxies at flux levels fainter than those detectable by the current instruments, and is expected to discover new objects. A large sample of stellar X-ray flares will also be detected and characterised. In the era of multi-messenger astronomy, EP has the potential of detecting the possible X-ray counterparts of gravitational wave events, neutrino sources, and ultra-high energy γ-ray and cosmic ray sources. EP is expected to help advance the studies of extreme objects/phenomena and their underlying physical processes revealed in the dynamic X-ray universe, as well as studies in other areas of X-ray astronomy.

astro-ph.HE↗

arXiv2025

ModelShield: Adaptive and Robust Watermark against Model Extraction Attack

Large language models (LLMs) demonstrate general intelligence across a variety of machine learning tasks, thereby enhancing the commercial value of their intellectual property (IP). To protect this IP, model owners typically allow user access only in a black-box manner, however, adversaries can still utilize model extraction attacks to steal the model intelligence encoded in model generation. Watermarking technology offers a promising solution for defending against such attacks by embedding unique identifiers into the model-generated content. However, existing watermarking methods often compromise the quality of generated content due to heuristic alterations and lack robust mechanisms to counteract adversarial strategies, thus limiting their practicality in real-world scenarios. In this paper, we introduce an adaptive and robust watermarking method (named ModelShield) to protect the IP of LLMs. Our method incorporates a self-watermarking mechanism that allows LLMs to autonomously insert watermarks into their generated content to avoid the degradation of model content. We also propose a robust watermark detection mechanism capable of effectively identifying watermark signals under the interference of varying adversarial strategies. Besides, ModelShield is a plug-and-play method that does not require additional model training, enhancing its applicability in LLM deployments. Extensive evaluations on two real-world datasets and three LLMs demonstrate that our method surpasses existing methods in terms of defense effectiveness and robustness while significantly reducing the degradation of watermarking on the model-generated content.

cs.CR↗

arXiv2025

Shifting-Merging: Secure, High-Capacity and Efficient Steganography via Large Language Models

In the face of escalating surveillance and censorship within the cyberspace, the sanctity of personal privacy has come under siege, necessitating the development of steganography, which offers a way to securely hide messages within innocent-looking texts. Previous methods alternate the texts to hide private massages, which is not secure. Large Language Models (LLMs) provide high-quality and explicit distribution, which is an available mathematical tool for secure steganography methods. However, existing attempts fail to achieve high capacity, time efficiency and correctness simultaneously, and their strongly coupling designs leave little room for refining them to achieve better performance. To provide a secure, high-capacity and efficient steganography method, we introduce ShiMer. Specifically, ShiMer pseudorandomly shifts the probability interval of the LLM's distribution to obtain a private distribution, and samples a token according to the private bits. ShiMer produced steganographic texts are indistinguishable in quality from the normal texts directly generated by the language model. To further enhance the capacity of ShiMer, we design a reordering algorithm to minimize the occurrence of interval splitting during decoding phase. Experimental results indicate that our method achieves the highest capacity and efficiency among existing secure steganography techniques.

cs.CR↗

arXiv2024

Semantic Steganography: A Framework for Robust and High-Capacity Information Hiding using Large Language Models

In the era of Large Language Models (LLMs), generative linguistic steganography has become a prevalent technique for hiding information within model-generated texts. However, traditional steganography methods struggle to effectively align steganographic texts with original model-generated texts due to the lower entropy of the predicted probability distribution of LLMs. This results in a decrease in embedding capacity and poses challenges for decoding stegos in real-world communication channels. To address these challenges, we propose a semantic steganography framework based on LLMs, which construct a semantic space and map secret messages onto this space using ontology-entity trees. This framework offers robustness and reliability for transmission in complex channels, as well as resistance to text rendering and word blocking. Additionally, the stegos generated by our framework are indistinguishable from the covers and achieve a higher embedding capacity compared to state-of-the-art steganography methods, while producing higher quality stegos.

cs.CR↗

arXiv2024

Training With "Paraphrasing the Original Text" Teaches LLM to Better Retrieve in Long-context Tasks

As Large Language Models (LLMs) continue to evolve, more are being designed to handle long-context inputs. Despite this advancement, most of them still face challenges in accurately handling long-context tasks, often showing the "lost in the middle" issue. We identify that insufficient retrieval capability is one of the important reasons for this issue. To tackle this challenge, we propose a novel approach to design training data for long-context tasks, aiming at augmenting LLMs' proficiency in extracting key information from long context. Specially, we incorporate an additional part named "paraphrasing the original text" when constructing the answer of training samples and then fine-tuning the model. Experimenting on LongBench and NaturalQuestions Multi-document-QA dataset with models of Llama and Qwen series, our method achieves an improvement of up to 8.48% and 4.48% in average scores, respectively, showing effectiveness in improving the model's performance on long-context tasks.

cs.CL↗