arXiv subjects

Jun Zhao

Publications and source records attributed to Jun Zhao.

At least 163 records · Page 9Linked to original sources

arXiv2025

JoyType: A Robust Design for Multilingual Visual Text Creation

Generating images with accurately represented text, especially in non-Latin languages, poses a significant challenge for diffusion models. Existing approaches, such as the integration of hint condition diagrams via auxiliary networks (e.g., ControlNet), have made strides towards addressing this issue. However, diffusion models often fall short in tasks requiring controlled text generation, such as specifying particular fonts or producing text in small fonts. In this paper, we introduce a novel approach for multilingual visual text creation, named JoyType, designed to maintain the font style of text during the image generation process. Our methodology begins with assembling a training dataset, JoyType-1M, comprising 1 million pairs of data. Each pair includes an image, its description, and glyph instructions corresponding to the font style within the image. We then developed a text control network, Font ControlNet, tasked with extracting font style information to steer the image generation. To further enhance our model's ability to maintain font style, notably in generating small-font text, we incorporated a multi-layer OCR-aware loss into the diffusion process. This enhancement allows JoyType to direct text rendering using low-level descriptors. Our evaluations, based on both visual and accuracy metrics, demonstrate that JoyType significantly outperforms existing state-of-the-art methods. Additionally, JoyType can function as a plugin, facilitating the creation of varied image styles in conjunction with other stable diffusion models on HuggingFace and CivitAI. Our project is open-sourced on https://jdh-algo.github.io/JoyType/.

cs.CV↗

arXiv2025

Revealing nanoscale structural phase separation in La$_{3}$Ni$_{2}$O$_{7-δ}$ single crystal via scanning near-field optical microscopy

The discovery of superconductivity in La3Ni2O7-$δ$ under high pressure,with an onset critical temperature around 80 K, has sparked significant interest in the superconducting phases of Ruddlesden-Popper nickelates, Lan+1NinO3n+1. While La4Ni3O10 exhibits nearly 100% superconductivity with Tc~30 K under high pressure, magnetic susceptibility studies on La3Ni2O7-$δ$, however, reveal a more complex picture, indicating either filamentary superconductivity or that approximately 50% of crystal phase becomes superconducting in polycrystalline samples. In this study, we employed scattering-type scanning near-field optical microscopy to visualize nanoscale structural phase separation in La3Ni2O7-$δ$, identifying enhanced optical conductivity with stripes approximately 183 nm wide. These stripes run diagonally with respect to the Ni-O-Ni bond directions in the a-b plane, ruling out the possibility that they arise from impurity phases, like the '1313', '214' or '4310' structures. The dark regions and bright stripes exhibit optical conductivities ~ 22% and 29% of gold's, respectively. Additionally, we find that the bright stripes constitute about 38% of the total field of view, while the remainder consists of dark regions and the transitional region between dark regions and bright stripes. Our results suggest that optical conductivity stripes originate from nanoscale structural phase separation. In contrast, La4Ni3O10 exhibits uniform and higher optical conductivity with no observable evidence of phase separation. Thus, our study represents a pioneering effort to directly image nanoscale phase separation in Lan+1NinO3n+1 nickelates. This observation could provide crucial insights into the factors that limit the superconducting volume fraction of La3Ni2O7-$δ$, highlighting SNOM as a powerful probe for exploring nanoscale low-energy physics in correlated quantum materials.

cond-mat.supr-con↗

arXiv2025

SAMRefiner: Taming Segment Anything Model for Universal Mask Refinement

In this paper, we explore a principal way to enhance the quality of widely pre-existing coarse masks, enabling them to serve as reliable training data for segmentation models to reduce the annotation cost. In contrast to prior refinement techniques that are tailored to specific models or tasks in a close-world manner, we propose SAMRefiner, a universal and efficient approach by adapting SAM to the mask refinement task. The core technique of our model is the noise-tolerant prompting scheme. Specifically, we introduce a multi-prompt excavation strategy to mine diverse input prompts for SAM (i.e., distance-guided points, context-aware elastic bounding boxes, and Gaussian-style masks) from initial coarse masks. These prompts can collaborate with each other to mitigate the effect of defects in coarse masks. In particular, considering the difficulty of SAM to handle the multi-object case in semantic segmentation, we introduce a split-then-merge (STM) pipeline. Additionally, we extend our method to SAMRefiner++ by introducing an additional IoU adaption step to further boost the performance of the generic SAMRefiner on the target dataset. This step is self-boosted and requires no additional annotation. The proposed framework is versatile and can flexibly cooperate with existing segmentation methods. We evaluate our mask framework on a wide range of benchmarks under different settings, demonstrating better accuracy and efficiency. SAMRefiner holds significant potential to expedite the evolution of refinement tools. Our code is available at https://github.com/linyq2117/SAMRefiner.

cs.CV↗

arXiv2025

FedSem: A Resource Allocation Scheme for Federated Learning Assisted Semantic Communication

Semantic communication (SemCom), regarded as the evolution of the traditional Shannon's communication model, stresses the transmission of semantic information instead of the data itself. Federated learning (FL), owing to its distributed learning and privacy-preserving properties, has received attention from both academia and industry. In this paper, we introduce a system that integrates FL and SemCom, which is called FedSem. We have also proposed an optimization problem related to resource allocation for this system. The objective of this problem is to minimize the energy consumption and delay of FL, as well as the transmission energy of SemCom, while maximizing the accuracy of the model trained through FL. The channel access scheme is Orthogonal Frequency-Division Multiple Access (OFDMA). The optimization variables include the binary (0-1) subcarrier allocation indicator, the transmission power of each device on specific subcarriers, the computational frequency of each participating device, and the compression rate for SemCom. To tackle this complex problem, we propose a resource allocation algorithm that decomposes the original problem into more tractable subproblems. By employing convex optimization techniques, we transform the non-convex problem into convex forms, ensuring tractability and solution effectiveness. Our approach includes a detailed analysis of time complexity and convergence, proving the practicality of the algorithm. Numerical experiments validate the effectiveness of our approach, showing superior performance of our algorithm in various scenarios compared to baseline methods. Hence, our solution is useful for enhancing the operational efficiency of FedSem systems, offering significant potential for real-world applications.

cs.DC↗

arXiv2025

Enhancing Large Language Models with Pseudo- and Multisource- Knowledge Graphs for Open-ended Question Answering

Mitigating the hallucinations of Large Language Models is a crucial task. Although some existing methods employ self-enhancement techniques, they fall short of effectively addressing unknown factual hallucinations. Meanwhile, Knowledge Graph (KG) enhancement approaches fail to address the generalization across different KG sources and the enhancement of open-ended answer questions simultaneously. To tackle these limitations, we propose a framework that combines Pseudo-Graph Generation and Atomic Knowledge Verification (PG\&AKV). Enhancement of open-ended question-answering begins with leveraging the Pseudo-Graph Generation to provide the related knowledge framework. Subsequently, Atomic Knowledge Verification utilizes atomic-level knowledge querying and verification to achieve generalizability under different KG sources. Compared to the baseline, this approach yields a minimum improvement of 11.5 in the ROUGE-L score for open-ended questions. For precise-answered questions, we observe a minimum accuracy improvement of 7.5%. Moreover, PG\&AKV also exhibits generalizability across different KG sources. Utilizing KG different from the question sources, PG\&AKV can even achieve at least a 3.5 % performance improvement. In summary, our results pave the way for enhancing LLMs by incorporating Pseudo- and Multisource-KGs, particularly in the filed of open-ended questions.

cs.CL↗

arXiv2025

MIRAGE: Evaluating and Explaining Inductive Reasoning Process in Language Models

Inductive reasoning is an essential capability for large language models (LLMs) to achieve higher intelligence, which requires the model to generalize rules from observed facts and then apply them to unseen examples. We present MIRAGE, a synthetic dataset that addresses the limitations of previous work, specifically the lack of comprehensive evaluation and flexible test data. In it, we evaluate LLMs' capabilities in both the inductive and deductive stages, allowing for flexible variation in input distribution, task scenario, and task difficulty to analyze the factors influencing LLMs' inductive reasoning. Based on these multi-faceted evaluations, we demonstrate that the LLM is a poor rule-based reasoner. In many cases, when conducting inductive reasoning, they do not rely on a correct rule to answer the unseen case. From the perspectives of different prompting methods, observation numbers, and task forms, models tend to consistently conduct correct deduction without correct inductive rules. Besides, we find that LLMs are good neighbor-based reasoners. In the inductive reasoning process, the model tends to focus on observed facts that are close to the current test example in feature space. By leveraging these similar examples, the model maintains strong inductive capabilities within a localized region, significantly improving its deductive performance.

cs.CL↗

arXiv2025

Capability Localization: Capabilities Can be Localized rather than Individual Knowledge

Large scale language models have achieved superior performance in tasks related to natural language processing, however, it is still unclear how model parameters affect performance improvement. Previous studies assumed that individual knowledge is stored in local parameters, and the storage form of individual knowledge is dispersed parameters, parameter layers, or parameter chains, which are not unified. We found through fidelity and reliability evaluation experiments that individual knowledge cannot be localized. Afterwards, we constructed a dataset for decoupling experiments and discovered the potential for localizing data commonalities. To further reveal this phenomenon, this paper proposes a Commonality Neuron Localization (CNL) method, which successfully locates commonality neurons and achieves a neuron overlap rate of 96.42% on the GSM8K dataset. Finally, we have demonstrated through cross data experiments that commonality neurons are a collection of capability neurons that possess the capability to enhance performance. Our code is available at https://github.com/nlpkeg/Capability-Neuron-Localization.

cs.CL↗

arXiv2025

Unraveling the origin of Kondo-like behavior in the 3$d$-electron heavy-fermion compound YFe$_{2}$Ge$_{2}$

The heavy fermion (HF) state of $d$-electron systems is of great current interest since it exhibits various exotic phases and phenomena that are reminiscent of the Kondo effect in $f$-electron HF systems. Here, we present a combined infrared spectroscopy and first-principles band structure calculation study of the $3d$-electron HF compound YFe$_2$Ge$_2$. The infrared response exhibits several charge-dynamical hallmarks of HF and a corresponding scaling behavior that resemble those of the $f$-electron HF systems. In particular, the low-temperature spectra reveal a dramatic narrowing of the Drude response along with the appearance of a hybridization gap ($Δ\sim$ 50 meV) and a strongly enhanced quasiparticle effective mass. Moreover, the temperature dependence of the infrared response indicates a crossover around $T^{\ast} \sim$ 100 K from a coherent state at low temperature to a quasi-incoherent one at high temperature. Despite of these striking similarities, our band structure calculations suggest that the mechanism underlying the HF behavior in YFe$_2$Ge$_2$ is distinct from the Kondo scenario of the $f$-electron HF compounds and even from that of the $d$-electron iron-arsenide superconductor KFe$_2$As$_2$. For the latter, the HF state is driven by orbital-selective correlations due to a strong Hund's coupling. Instead, for YFe$_2$Ge$_2$ the HF behavior originates from the band flatness near the Fermi level induced by the combined effects of kinetic frustration from a destructive interference between the direct Fe-Fe and indirect Fe-Ge-Fe hoppings, band hybridization involving Fe $3d$ and Y $4d$ electrons, and electron correlations. This highlights that rather different mechanisms can be at the heart of the HF state in $d$-electron systems.

cond-mat.str-el↗

arXiv2025

Knowledge Localization: Mission Not Accomplished? Enter Query Localization!

Large language models (LLMs) store extensive factual knowledge, but the mechanisms behind how they store and express this knowledge remain unclear. The Knowledge Neuron (KN) thesis is a prominent theory for explaining these mechanisms. This theory is based on the Knowledge Localization (KL) assumption, which suggests that a fact can be localized to a few knowledge storage units, namely knowledge neurons. However, this assumption has two limitations: first, it may be too rigid regarding knowledge storage, and second, it neglects the role of the attention module in knowledge expression. In this paper, we first re-examine the KL assumption and demonstrate that its limitations do indeed exist. To address these, we then present two new findings, each targeting one of the limitations: one focusing on knowledge storage and the other on knowledge expression. We summarize these findings as \textbf{Query Localization} (QL) assumption and argue that the KL assumption can be viewed as a simplification of the QL assumption. Based on QL assumption, we further propose the Consistency-Aware KN modification method, which improves the performance of knowledge modification, further validating our new assumption. We conduct 39 sets of experiments, along with additional visualization experiments, to rigorously confirm our conclusions. Code is available at https://github.com/heng840/KnowledgeLocalization.

cs.CL↗

arXiv2025

The Knowledge Microscope: Features as Better Analytical Lenses than Neurons

Previous studies primarily utilize MLP neurons as units of analysis for understanding the mechanisms of factual knowledge in Language Models (LMs); however, neurons suffer from polysemanticity, leading to limited knowledge expression and poor interpretability. In this paper, we first conduct preliminary experiments to validate that Sparse Autoencoders (SAE) can effectively decompose neurons into features, which serve as alternative analytical units. With this established, our core findings reveal three key advantages of features over neurons: (1) Features exhibit stronger influence on knowledge expression and superior interpretability. (2) Features demonstrate enhanced monosemanticity, showing distinct activation patterns between related and unrelated facts. (3) Features achieve better privacy protection than neurons, demonstrated through our proposed FeatureEdit method, which significantly outperforms existing neuron-based approaches in erasing privacy-sensitive information from LMs.Code and dataset will be available.

cs.CL↗

arXiv2025

Knowledge in Superposition: Unveiling the Failures of Lifelong Knowledge Editing for Large Language Models

Knowledge editing aims to update outdated or incorrect knowledge in large language models (LLMs). However, current knowledge editing methods have limited scalability for lifelong editing. This study explores the fundamental reason why knowledge editing fails in lifelong editing. We begin with the closed-form solution derived from linear associative memory, which underpins state-of-the-art knowledge editing methods. We extend the solution from single editing to lifelong editing, and through rigorous mathematical derivation, identify an interference term in the final solution, suggesting that editing knowledge may impact irrelevant knowledge. Further analysis of the interference term reveals a close relationship with superposition between knowledge representations. When knowledge superposition does not exist in language models, the interference term vanishes, allowing for lossless knowledge editing. Experiments across numerous language models reveal that knowledge superposition is universal, exhibiting high kurtosis, zero mean, and heavy-tailed distributions with clear scaling laws. Ultimately, by combining theory and experiments, we demonstrate that knowledge superposition is the fundamental reason for the failure of lifelong editing. Moreover, this is the first study to investigate knowledge editing from the perspective of superposition and provides a comprehensive observation of superposition across numerous real-world language models. Code available at https://github.com/ChenhuiHu/knowledge_in_superposition.

cs.CL↗

arXiv2025

Continual Reinforcement Learning for Digital Twin Synchronization Optimization

This article investigates the adaptive resource allocation scheme for digital twin (DT) synchronization optimization over dynamic wireless networks. In our considered model, a base station (BS) continuously collects factory physical object state data from wireless devices to build a real-time virtual DT system for factory event analysis. Due to continuous data transmission, maintaining DT synchronization must use extensive wireless resources. To address this issue, a subset of devices is selected to transmit their sensing data, and resource block (RB) allocation is optimized. This problem is formulated as a constrained Markov process (CMDP) problem that minimizes the long-term mismatch between the physical and virtual systems. To solve this CMDP, we first transform the problem into a dual problem that refines RB constraint impacts on device scheduling strategies. We then propose a continual reinforcement learning (CRL) algorithm to solve the dual problem. The CRL algorithm learns a stable policy across historical experiences for quick adaptation to dynamics in physical states and network capacity. Simulation results show that the CRL can adapt quickly to network capacity changes and reduce normalized root mean square error (NRMSE) between physical and virtual states by up to 55.2%, using the same RB number as traditional methods.

cs.NI↗

arXiv2025

Isotropic superconductivity in pressurized trilayer nickelate La4Ni3O10

Evidence of superconductivity (SC) has recently been reported in pressurized La3Ni2O7 and La4Ni3O10, providing a new platform to explore high-temperature superconductivity. However, while zero resistance state has been observed, experimental characterization of the superconducting properties of pressurized nickelates is still limited and experimentally challenging. Here, we present the first full temperature dependence of the upper critical field Hc2 measurement in La4Ni3O10 single crystal, achieved by combining high magnetic field and high-pressure techniques. Remarkably, the Hc2 of La4Ni3O10 is nearly isotropic, with the anisotropic parameter monotonically increasing from 1.4 near Tc to 1 at lower temperatures. By analyzing the Hc2 using the two-band model, we uncover that the anisotropic diffusivity of the bands, primarily originating from d(z2 ) and d(x2-y2 ) orbitals, is well compensated, resulting in an unusually isotropic superconducting state. These findings provide critical experimental evidence that underscores the significant role of the d(z2 ) orbital in enabling superconductivity in pressurized Ruddlesden-Popper nickelates.

cond-mat.supr-con↗

arXiv2025

GATE: Graph-based Adaptive Tool Evolution Across Diverse Tasks

Large Language Models (LLMs) have shown great promise in tool-making, yet existing frameworks often struggle to efficiently construct reliable toolsets and are limited to single-task settings. To address these challenges, we propose GATE (Graph-based Adaptive Tool Evolution), an adaptive framework that dynamically constructs and evolves a hierarchical graph of reusable tools across multiple scenarios. We evaluate GATE on open-ended tasks (Minecraft), agent-based tasks (TextCraft, DABench), and code generation tasks (MATH, Date, TabMWP). Our results show that GATE achieves up to 4.3x faster milestone completion in Minecraft compared to the previous SOTA, and provides an average improvement of 9.23% over existing tool-making methods in code generation tasks and 10.03% in agent tasks. GATE demonstrates the power of adaptive evolution, balancing tool quantity, complexity, and functionality while maintaining high efficiency. Code and data are available at \url{https://github.com/ayanami2003/GATE}.

cs.CL↗

arXiv2025

From Instance Training to Instruction Learning: Task Adapters Generation from Instructions

Large language models (LLMs) have acquired the ability to solve general tasks by utilizing instruction finetuning (IFT). However, IFT still relies heavily on instance training of extensive task data, which greatly limits the adaptability of LLMs to real-world scenarios where labeled task instances are scarce and broader task generalization becomes paramount. Contrary to LLMs, humans acquire skills and complete tasks not merely through repeated practice but also by understanding and following instructional guidelines. This paper is dedicated to simulating human learning to address the shortcomings of instance training, focusing on instruction learning to enhance cross-task generalization. Within this context, we introduce Task Adapters Generation from Instructions (TAGI), which automatically constructs the task-specific model in a parameter generation manner based on the given task instructions without retraining for unseen tasks. Specifically, we utilize knowledge distillation to enhance the consistency between TAGI developed through Learning with Instruction and task-specific models developed through Training with Instance, by aligning the labels, output logits, and adapter parameters between them. TAGI is endowed with cross-task generalization capabilities through a two-stage training process that includes hypernetwork pretraining and finetuning. We evaluate TAGI on the Super-Natural Instructions and P3 datasets. The experimental results demonstrate that TAGI can match or even outperform traditional meta-trained models and other hypernetwork models, while significantly reducing computational requirements.

cs.CL↗

arXiv2025

Neural-Symbolic Collaborative Distillation: Advancing Small Language Models for Complex Reasoning Tasks

In this paper, we propose $\textbf{Ne}$ural-$\textbf{Sy}$mbolic $\textbf{C}$ollaborative $\textbf{D}$istillation ($\textbf{NesyCD}$), a novel knowledge distillation method for learning the complex reasoning abilities of Large Language Models (LLMs, e.g., \textgreater 13B). We argue that complex reasoning tasks are difficult for Small Language Models (SLMs, e.g., $\leq$ 7B), as these tasks demand not only general cognitive abilities but also specialized knowledge, which is often sparse and difficult for these neural-based SLMs to effectively capture. Therefore, NesyCD distills the general capabilities and specialized knowledge in LLMs using different manners. On the one hand, we distill only general abilities from teacher LLMs into the student SLMs of parameterized neural networks. On the other hand, for the specialized abilities and uncommon knowledge of a complex reasoning task, we employ a symbolic knowledge distillation approach to obtain and store the specialized knowledge within a symbolic knowledge base (KB). By decoupling general and specialized capabilities, the proposed NesyCD can achieve superior performance cost-effectively, utilizing smaller models and blending parameterized neural networks with symbolic KB. Moreover, the specialized KB generalizes well and is comprehended and manipulated by humans. Our experiments show that NesyCD significantly boosts SLMs' complex reasoning performance on in-domain (BBH, GSM8K) and out-of-domain (AGIEval, ARC) datasets. Notably, our approach enabled the LLaMA3-8B and Qwen2-7B to surpass GPT-3.5-turbo in performance and come close to matching LLaMA3-70B, despite the latter having nine times more parameters. Our code will be available at https://github.com/Xnhyacinth/NesyCD.

cs.CL↗

arXiv2025

Does RAG Really Perform Bad For Long-Context Processing?

The efficient processing of long context poses a serious challenge for large language models (LLMs). Recently, retrieval-augmented generation (RAG) has emerged as a promising strategy for this problem, as it enables LLMs to make selective use of the long context for efficient computation. However, existing RAG approaches lag behind other long-context processing methods due to inherent limitations on inaccurate retrieval and fragmented contexts. To address these challenges, we introduce RetroLM, a novel RAG framework for long-context processing. Unlike traditional methods, RetroLM employs KV-level retrieval augmentation, where it partitions the LLM's KV cache into contiguous pages and retrieves the most crucial ones for efficient computation. This approach enhances robustness to retrieval inaccuracy, facilitates effective utilization of fragmented contexts, and saves the cost from repeated computation. Building on this framework, we further develop a specialized retriever for precise retrieval of critical pages and conduct unsupervised post-training to optimize the model's ability to leverage retrieved information. We conduct comprehensive evaluations with a variety of benchmarks, including LongBench, InfiniteBench, and RULER, where RetroLM significantly outperforms existing long-context LLMs and efficient long-context processing methods, particularly in tasks requiring intensive reasoning or extremely long-context comprehension.

cs.CL↗

arXiv2025

DATA: Decomposed Attention-based Task Adaptation for Rehearsal-Free Continual Learning

Continual learning (CL) is essential for Large Language Models (LLMs) to adapt to evolving real-world demands, yet they are susceptible to catastrophic forgetting (CF). While traditional CF solutions rely on expensive data rehearsal, recent rehearsal-free methods employ model-based and regularization-based strategies to address this issue. However, these approaches often neglect the model's plasticity, which is crucial to achieving optimal performance on newly learned tasks. Consequently, a key challenge in CL is striking a balance between preserving plasticity and mitigating CF. To tackle this challenge, we propose the $\textbf{D}$ecomposed $\textbf{A}$ttention-based $\textbf{T}$ask $\textbf{A}$daptation (DATA), which explicitly decouples and learns both task-specific and task-shared knowledge using high-rank and low-rank task adapters (e.g., LoRAs). For new tasks, DATA dynamically adjusts the weights of adapters of different ranks based on their relevance and distinction from previous tasks, allowing the model to acquire new task-specific skills while effectively retaining previously learned knowledge. Specifically, we implement a decomposed component weighting strategy comprising learnable components that collectively generate attention-based weights, allowing the model to integrate and utilize diverse knowledge from each DATA. Extensive experiments on three widely used benchmarks demonstrate that our proposed method achieves state-of-the-art performance. Notably, our approach significantly enhances model plasticity and mitigates CF by extending learnable components and employing stochastic restoration during training iterations.

cs.LG↗