arXiv subjects

Lin Sun

Publications and source records attributed to Lin Sun.

At least 37 records · Page 2Linked to original sources

arXiv2026

BEAR: Budgeted Evidence Allocation for Multi-Document Reasoning

We argue that multi-document reasoning is constrained not only by how much text a model can read, but also by how limited query-time evidence budget is allocated across documents and semantic granularities. Full-context inference exposes the model to broad evidence non-selectively and at high per-query cost, while flat chunk retrieval often returns locally relevant passages that are weakly organized for cross-document synthesis. We present \textbf{BEAR}, a framework for structured evidence allocation that builds hierarchical semantic indices offline and performs coarse-to-fine evidence access at query time through complementary \emph{exploration} and \emph{recovery} paths. This coarse-to-fine design can be viewed as structured evidence allocation under a fixed evidence-context budget. Across synthetic and real-world benchmarks, BEAR performs particularly strongly on DragonBall, remains competitive with strong retrieval-based baselines on HotpotQA, and yields the best retrieval-based result on 2Wiki under our evaluated protocol, while operating under substantially smaller \emph{query-time evidence budgets} than the reported long-context references. Additional analyses suggest that the gains are associated with hierarchy as an allocation substrate together with complementary exploration and recovery, rather than semantic chunking alone.

cs.CL↗

arXiv2026

MemAudit: Post-hoc Auditing of Poisoned Agent Memory via Causal Attribution and Structural Anomaly Detection

Large language model agents increasingly rely on persistent memory to store past interactions, retrieve relevant demonstrations, and improve long-horizon task execution. However, this memory mechanism also creates a practical security vulnerability: an adversarial user may inject malicious records into the agent's memory through ordinary interaction, and these records can later be retrieved to steer the agent's reasoning and actions. Existing defenses primarily focus on online intervention, such as prompt filtering or output blocking, but they do not address the post-hoc question of which stored memories are responsible after harmful behavior has already been observed. We propose \textbf{MemAudit}, a post-hoc causal memory auditing framework for memory-augmented LLM agents. The framework combines two complementary signals: (1) a counterfactual memory influence score that measures each memory's causal contribution to harmful outputs, and (2) a memory consistency graph that identifies structurally anomalous memories within the broader memory store. We evaluate MemAudit against MINJA, a query-only memory injection attack in which malicious records are generated and stored through normal agent interactions rather than direct memory-bank modification. Across both QA and reasoning-agent settings, MemAudit substantially reduces attack success rates under realistic post-hoc auditing scenarios. The results show that QA attack success is reduced from $70\%$ to $0\%$, while RAP attack success drops from $83.3\%$ to $0\%$.

cs.AI↗

arXiv2026

Regularizations for shock and rarefaction waves in the perturbed solitons of the KP equation

Using an asymptotic perturbation method, we study the initial value problem for the KP equation with initial data consisting of parts of exact line-soliton solutions. We consider a slow modulation of the soliton parameters, described by a dynamical system obtained via the perturbation method. {The dynamical system is given by a $2$-component quasi-linear system.} In particular, we show that a singular solution (\emph{shock wave}) {of the system} leads to the generation of a new soliton as a result of the resonant interaction of solitons. We also show that a regular solution corresponding to a rarefaction wave {of the system} can be described by a parabola (which we call a \emph{parabolic soliton}). We then perform numerical simulations of the initial value problem and show that they are in excellent agreement with the results obtained by the perturbation method.

nlin.PS↗

arXiv2026

TinyR1-32B-Preview: Boosting Accuracy with Branch-Merge Distillation

The challenge of reducing the size of Large Language Models (LLMs) while maintaining their performance has gained significant attention. However, existing methods, such as model distillation and transfer learning, often fail to achieve high accuracy. To address this limitation, we introduce the Branch-Merge distillation approach, which enhances model compression through two phases: (1) the Branch Phase, where knowledge from a large teacher model is \textit{selectively distilled} into specialized student models via domain-specific supervised fine-tuning (SFT); And (2) the Merge Phase, where these student models are merged to enable cross-domain knowledge transfer and improve generalization. We validate our distillation approach using DeepSeek-R1 as the teacher and DeepSeek-R1-Distill-Qwen-32B as the student. The resulting merged model, TinyR1-32B-Preview, outperforms its counterpart DeepSeek-R1-Distill-Qwen-32B across multiple benchmarks, including Mathematics (+5.5 points), Coding (+4.4 points) and Science (+2.9 points), while achieving near-equal performance to DeepSeek-R1 on AIME 2024. The Branch-Merge distillation approach provides a scalable solution for creating smaller, high-performing LLMs with reduced computational cost and time.

cs.CL↗

arXiv2026

When Good OCR Is Not Enough: Benchmarking OCR Robustness for Retrieval-Augmented Generation

Industrial Retrieval-Augmented Generation (RAG) systems depend on optical character recognition (OCR) to transform visual documents into text. Existing OCR benchmarks rely on character-level metrics, which inadequately measure downstream RAG effectiveness under real-world conditions. We introduce an OCR benchmark for industrial RAG systems covering 11 challenging document types, including extreme layouts, high-resolution pages, complex or watermarked backgrounds, historical documents with non-standard reading orders, visually decorated text, and documents containing tables and mathematical formulas. Evaluating recent SOTA OCR models under a controlled OCR-first RAG pipeline shows clear performance degradation on realistic industrial documents despite strong conventional benchmark scores. We find that high OCR accuracy does not necessarily translate into strong downstream RAG performance: structural and semantic errors can cause substantial retrieval failures even when WER/CER remains low. Further analysis shows that this mismatch is category-dependent, arises through both retrieval-side and downstream generation-side failures, and remains stable across representative OCR-first pipeline choices. The benchmark is publicly available at https://github.com/Qihoo360/InduOCRBench.

cs.CV↗

arXiv2026

Thinking with Reasoning Skills: Fewer Tokens, More Accuracy

Reasoning LLMs often spend substantial tokens on long intermediate reasoning traces (e.g., chain-of-thought) when solving new problems. We propose to summarize and store reusable reasoning skills distilled from extensive deliberation and trial-and-error exploration, and to retrieve these skills at inference time to guide future reasoning. Unlike the prevailing \emph{reasoning from scratch} paradigm, our approach first recalls relevant skills for each query, helping the model avoid redundant detours and focus on effective solution paths. We evaluate our method on coding and mathematical reasoning tasks, and find that it significantly reduces reasoning tokens while improving overall performance. The resulting lower per-request cost indicates strong practical and economic potential for real-world deployment.

cs.AI↗

arXiv2026

Petabit-per-second Random Number Generation

Physical random number generators based on chaotic microcombs, with their complex nonlinear dynamics and multi-channel parallel capability, have attracted considerable research attention. However, key technical challenges for chaotic microcombs are the high correlation between symmetric teeth and the low bandwidth of single-channel teeth, which seriously affect the speed and scalability of random number generation. We experimentally demonstrate a petabit-per-second (Pbit/s) parallel random number generation system based on intensity chaotic modulation and Rayleigh scattering. Through intensity modulation, the effective bandwidth of the single-channel entropy source is increased from 440MHz to 27.6GHz. Crucially, Rayleigh scattering further contributes through the random superposition of backscattered light, which introduces unpredictable fluctuations in intensity, phase, and polarization. This randomness suppresses inter-channel correlation among parallel entropy sources to ~0.02, ensuring their orthogonality. Moreover, by employing polarization-diverse coherent detection on a single-channel, four new low correlated sub-channels are extracted: X-/Y- intensity and phase. We achieve a single-channel bit rate of 14.336 Tbit/s and a total bit rate of 1.032 Pbit/s (over 72 parallel channels) with offline post-processing, representing the highest post-processing record reported in both the single-channel and the total system. Moreover, our scheme based on a single chaotic microcomb and fiber scattering link show fundamentally scalable. The total bit rate can be significantly pushed beyond the Pbit/s level by further expanding the usable comb channel and/or by deploying multiple fiber scattering links in parallel, paving a practical path toward higher throughput regimes.

physics.optics↗

arXiv2026

Spectral study of the pseudogap in unitary Fermi gases

The existence of a pseudogap in unitary Fermi gases has recently been established and measured experimentally [Li et al., Nature 626, 288 (2024)]. This lends strong support for the pairing origin as the mechanism of the pseudogap in Fermi superfluids. Here we present a spectral study of unitary Fermi gases, and show how the data can be understood quantitatively, when compared with theoretically calculated momentum-resolved rf or microwave spectra, and the pseudogap extracted from the spectra. We use an iterative treatment of the fermion self energy and hence the spectral function, beyond previous pseudogap approximation, based on a pairing fluctuation theory that incorporates both particle-particle and particle-hole T matrices, with self-consistent self energy feedback. Our results not only provide a microscopic explanation of the experimental data but also strengthen the support for both the pairing-induced pseudogap physics and the pairing fluctuation theory of Fermi superfluidity.

cond-mat.quant-gas↗

arXiv2026

Rf spectra and pseudogap in ultracold Fermi gases across the BCS-BEC crossover from pairing fluctuation theory

The pseudogap phenomenon is a hallmark of strongly interacting Fermi systems, from high-temperature superconductors to ultracold atomic gases, yet its precise origin remains debated. Here we calculate the spectral function and rf spectra of ultracold atomic gases across the BCS-BEC crossover to quantitatively investigate the pairing mechanism of the pseudogap. We advance our pairing fluctuation theory by incorporating particle-hole fluctuations, which renormalize the effective interaction in the particle-particle channel. To achieve quantitative accuracy, we employ a full numerical convolution for the pair susceptibility and self-energy, moving beyond previous analytic pseudogap approximations. This convolution approach automatically captures two critical effects: (i) the full spectral broadening of fermions due to finite pair lifetime, and (ii) the previously neglected pair-hole scattering effect, which manifests as a substantial Hartree energy. We calculate the spectral function, and use rf spectral intensity maps and energy distribution curves to determine the quasiparticle dispersion. From these, we extract the pseudogap $Δ$, Hartree energy, and chemical potential, mapping their evolution across the crossover. Our results show that the pseudogap emerges continuously as the system moves from the BCS regime toward BEC. Furthermore, the pair spectral function reveals that pairs become diffusive at energies above 2$Δ$, indicating that the pair lifetime is governed by virtual binding and unbinding processes. Our calculations achieve quantitative agreement with recent experiments across the BCS-BEC crossover, including at unitarity, providing strong support for a pairing-based origin of the pseudogap as described by our pairing fluctuation theory.

cond-mat.quant-gas↗

arXiv2026

Effects of particle-hole fluctuations on the superfluid transition in two-dimensional atomic Fermi gases

Proper treatment of the many-body interactions is of paramount importance in our understanding of strongly correlated systems. Here we investigate the effects of particle-hole fluctuations on the Berezinskii-Kosterlitz-Thouless (BKT) transition in two-dimensional Fermi gases throughout the entire BCS-BEC crossover. We include self-consistently in the self energy treatment the entire particle-hole $T$ matrix, which constitutes a renormalization of the bare interaction that appears in the particle-particle scattering $T$ matrix, leading to a screening of the pairing interaction and hence a dramatic reduction of the pairing gap and the transition temperature. The BKT transition temperature $T_\text{BKT}$ is determined by the critical phase space density, for which the pair density and pair mass are determined using a pairing fluctuation theory, which accommodates self-consistently the important self-energy feedback in the treatment of finite-momentum pairing fluctuations. The screening strength varies continuously from its maximum in the BCS limit to essentially zero in BEC limit. In the unitary regime, it leads to an interaction-dependent shift of $T_\text{BKT}$ towards the BEC regime. This shift is crucial in an attempt to explain experimental data quantitatively, which often depends on the interaction strength. Our findings are consistent with available experimental results in the unitary and BEC regimes and with quantum Monte Carlo simulations in the BCS and unitary regimes.

cond-mat.quant-gas↗

arXiv2026

Beyond Parameter Arithmetic: Sparse Complementary Fusion for Distribution-Aware Model Merging

Model merging has emerged as a promising paradigm for composing the capabilities of large language models by directly operating in weight space, enabling the integration of specialized models without costly retraining. However, existing merging methods largely rely on parameter-space heuristics, which often introduce severe interference, leading to degraded generalization and unstable generation behaviors such as repetition and incoherent outputs. In this work, we propose Sparse Complementary Fusion with reverse KL (SCF-RKL), a novel model merging framework that explicitly controls functional interference through sparse, distribution-aware updates. Instead of assuming linear additivity in parameter space, SCF-RKL measures the functional divergence between models using reverse Kullback-Leibler divergence and selectively incorporates complementary parameters. This mode-seeking, sparsity-inducing design effectively preserves stable representations while integrating new capabilities. We evaluate SCF-RKL across a wide range of model scales and architectures, covering both reasoning-focused and instruction-tuned models. Extensive experiments on 24 benchmarks spanning advanced reasoning, general reasoning and knowledge, instruction following, and safety demonstrate, vision classification that SCF-RKL consistently outperforms existing model merging methods while maintaining strong generalization and generation stability.

cs.AI↗

arXiv2026

Beyond Static Alignment: Hierarchical Policy Control for LLM Safety via Risk-Aware Chain-of-Thought

Large Language Models (LLMs) face a fundamental safety-helpfulness trade-off due to static, one-size-fits-all safety policies that lack runtime controllabilityxf, making it difficult to tailor responses to diverse application needs. %As a result, models may over-refuse benign requests or under-constrain harmful ones. We present \textbf{PACT} (Prompt-configured Action via Chain-of-Thought), a framework for dynamic safety control through explicit, risk-aware reasoning. PACT operates under a hierarchical policy architecture: a non-overridable global safety policy establishes immutable boundaries for critical risks (e.g., child safety, violent extremism), while user-defined policies can introduce domain-specific (non-global) risk categories and specify label-to-action behaviors to improve utility in real-world deployment settings. The framework decomposes safety decisions into structured Classify$\rightarrow$Act paths that route queries to the appropriate action (comply, guide, or reject) and render the decision-making process transparent. Extensive experiments demonstrate that PACT achieves near state-of-the-art safety performance under global policy evaluation while attaining the best controllability under user-specific policy evaluation, effectively mitigating the safety-helpfulness trade-off. We will release the PACT model suite, training data, and evaluation protocols to facilitate reproducible research in controllable safety alignment.

cs.CL↗

arXiv2026

MemoryVLA: Perceptual-Cognitive Memory in Vision-Language-Action Models for Robotic Manipulation

Temporal context is essential for robotic manipulation because such tasks are inherently non-Markovian, yet mainstream VLA models typically overlook it and struggle with long-horizon, temporally dependent tasks. Cognitive science suggests that humans rely on working memory to buffer short-lived representations for immediate control, while the hippocampal system preserves verbatim episodic details and semantic gist of past experience for long-term memory. Inspired by these mechanisms, we propose MemoryVLA, a Cognition-Memory-Action framework for long-horizon robotic manipulation. A pretrained VLM encodes the observation into perceptual and cognitive tokens that form working memory, while a Perceptual-Cognitive Memory Bank stores low-level details and high-level semantics consolidated from it. Working memory retrieves decision-relevant entries from the bank, adaptively fuses them with current tokens, and updates the bank by merging redundancies. Using these tokens, a memory-conditioned diffusion action expert yields temporally aware action sequences. We evaluate MemoryVLA on 150+ simulation and real-world tasks across three robots. On SimplerEnv-Bridge, Fractal, LIBERO-5 suites and Mikasa-Robo, it achieves 71.9%, 72.7%, 96.5%, and 41.2% success rates, respectively, all outperforming state-of-the-art baselines CogACT and pi-0, with a notable +14.6 gain on Bridge and +11.8 gain on Mikasa-Robo. On 12 real-world tasks spanning general skills and long-horizon temporal dependencies, MemoryVLA achieves 84.0% success rate, with long-horizon tasks showing a +26 improvement over state-of-the-art baseline. Project Page: https://shihao1895.github.io/MemoryVLA

cs.RO↗

arXiv2026

TriPlay-RL: Tri-Role Self-Play Reinforcement Learning for LLM Safety Alignment

In recent years, safety risks associated with large language models have become increasingly prominent, highlighting the urgent need to mitigate the generation of toxic and harmful content. The mainstream paradigm for LLM safety alignment typically adopts a collaborative framework involving three roles: an attacker for adversarial prompt generation, a defender for safety defense, and an evaluator for response assessment. In this paper, we propose a closed-loop reinforcement learning framework called TriPlay-RL that enables iterative and co-improving collaboration among three roles with near-zero manual annotation. Experimental results show that the attacker preserves high output diversity while achieving a 20%-50% improvement in adversarial effectiveness; the defender attains 10%-30% gains in safety performance without degrading general reasoning capability; and the evaluator continuously refines its fine-grained judgment ability through iterations, accurately distinguishing unsafe responses, simple refusals, and useful guidance. Overall, our framework establishes an efficient and scalable paradigm for LLM safety alignment, enabling continuous co-evolution within a unified learning loop.

cs.LG↗

arXiv2026

Efficient Switchable Safety Control in LLMs via Magic-Token-Guided Co-Training

Current methods for content safety in Large Language Models (LLMs), such as Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF), often rely on multi-stage training pipelines and lack fine-grained, post-deployment controllability. To address these limitations, we propose a unified co-training framework that efficiently integrates multiple safety behaviors: positive (lawful/prosocial), negative (unfiltered/risk-prone) and rejective (refusal-oriented/conservative) within a single SFT stage. Notably, each behavior is dynamically activated via a simple system-level instruction, or magic token, enabling stealthy and efficient behavioral switching at inference time. This flexibility supports diverse deployment scenarios, such as positive for safe user interaction, negative for internal red-teaming, and rejective for context-aware refusals triggered by upstream moderation signals. This co-training strategy induces a distinct Safety Alignment Margin in the output space, characterized by well-separated response distributions corresponding to each safety mode. The existence of this margin provides empirical evidence for the model's safety robustness and enables unprecedented fine-grained control. Experiments show that our method matches the safety alignment quality of SFT+DPO, with our 8B model notably surpassing DeepSeek-R1 (671B) in safety performance, while significantly reducing both training complexity and deployment costs. This work presents a scalable, efficient, and highly controllable solution for LLM content safety.

cs.CL↗

arXiv2025

On the maximum spectral radius of planar graphs

This paper investigates the maximum spectral radius of planar graphs with concrete fixed number of vertices, providing some tight bounds on the maximum spectral radius of general planar graph resorting to its order, and confirming that among all planar graphs containing dominating vertex with concrete fixed order $n \geq 48$, the join of $P_{2}$ and $P_{n-2}$ attains the maximum spectral radius.

math.CO↗

arXiv2025

Dexbotic: Open-Source Vision-Language-Action Toolbox

In this paper, we present Dexbotic, an open-source Vision-Language-Action (VLA) model toolbox based on PyTorch. It aims to provide a one-stop VLA research service for professionals in the field of embodied intelligence. It offers a codebase that supports multiple mainstream VLA policies simultaneously, allowing users to reproduce various VLA methods with just a single environment setup. The toolbox is experiment-centric, where the users can quickly develop new VLA experiments by simply modifying the Exp script. Moreover, we provide much stronger pretrained models to achieve great performance improvements for state-of-the-art VLA policies. Dexbotic will continuously update to include more of the latest pre-trained foundation models and cutting-edge VLA models in the industry.

cs.RO↗

arXiv2025

Large Language Models Badly Generalize across Option Length, Problem Types, and Irrelevant Noun Replacements

In this paper, we propose a ``Generalization Stress Test" to assess Large Language Models' (LLMs) generalization ability under slight and controlled perturbations, including option length, problem types, and irrelevant noun replacements. We achieve novel and significant findings that, despite high benchmark scores, LLMs exhibit severe accuracy drops and unexpected biases (e.g., preference for longer distractors) when faced with these minor but content-preserving modifications. For example, Qwen 2.5 1.5B's MMLU score rises from 60 to 89 and drops from 89 to 36 when option lengths are changed without altering the question. Even GPT4o experiences a 25-point accuracy loss when problem types are changed, with a 6-point drop across all three modification categories. These analyses suggest that LLMs rely heavily on superficial cues rather than forming robust, abstract representations that generalize across formats, lexical variations, and irrelevant content shifts.

cs.CL↗