arXiv subjects

Xing Li

Publications and source records attributed to Xing Li.

At least 37 records · Page 2Linked to original sources

arXiv2026

Surrogate Modeling with Low-Rank Function Representation for Electromagnetic Simulation

High-fidelity electromagnetic (EM) simulations are indispensable for the design of microwave and wave devices, yet repeated full-wave evaluations over high-dimensional design spaces are often computationally prohibitive. While neural surrogates can amortize this cost, learning high-dimensional EM response mappings remains difficult under limited simulation budgets due to strong and heterogeneous parameter couplings. In this work, we introduce low-rank tensor function representations as a principled surrogate modeling paradigm for EM problems and provide a systematic study of representative low-rank formats, including Tucker-style low-rank tensor function representation (LRTFR) as well as neural functional tensor-train (TT) and tensor-ring (TR) baselines. Building on these insights, we propose a pairwise low-rank tensor network (PLRNet) that uses learnable pairwise interaction factors over compact coordinate-wise embeddings. Experiments on representative EM surrogate tasks demonstrate that the proposed framework achieves a more favorable overall trade-off between accuracy, robustness, and parameter efficiency, with stable optimization in high-dimensional regimes.

cs.CE↗

arXiv2026

Unleashing Low-Bit Inference on Ascend NPUs: A Comprehensive Evaluation of HiFloat Formats

As LLMs scale, low-bit floating-point formats like MXFP and NVFP4 offer new opportunities for precision and efficiency. In this work, we evaluate HiFloat (HiF8 and HiF4), a family of formats tailored for Ascend NPUs. Through rigorous comparison across weight-activation and KV-cache tasks, we provide three key insights: (1) INT8 suits narrow-range data, while floating-point formats excel with high-variance data; (2) in 4-bit regimes, HiF4's hierarchical scaling prevents the accuracy collapse seen in integer formats; and (3) HiFloat is fully compatible with state-of-the-art post-training quantization frameworks. Overall, HiFloat provides a solution for high-efficiency LLM inference on NPUs.

cs.CL↗

arXiv2026

Towards Efficient Agents: A Co-Design of Inference Architecture and System

The rapid development of large language model (LLM)-based agents has unlocked new possibilities for autonomous multi-turn reasoning and tool-augmented decision-making. However, their real-world deployment is hindered by severe inefficiencies that arise not from isolated model inference, but from the systemic latency accumulated across reasoning loops, context growth, and heterogeneous tool interactions. This paper presents AgentInfer, a unified framework for end-to-end agent acceleration that bridges inference optimization and architectural design. We decompose the problem into four synergistic components: AgentCollab, a hierarchical dual-model reasoning framework that balances large- and small-model usage through dynamic role assignment; AgentSched, a cache-aware hybrid scheduler that minimizes latency under heterogeneous request patterns; AgentSAM, a suffix-automaton-based speculative decoding method that reuses multi-session semantic memory to achieve low-overhead inference acceleration; and AgentCompress, a semantic compression mechanism that asynchronously distills and reorganizes agent memory without disrupting ongoing reasoning. Together, these modules form a Self-Evolution Engine capable of sustaining efficiency and cognitive stability throughout long-horizon reasoning tasks. Experiments on the BrowseComp-zh and DeepDiver benchmarks demonstrate that through the synergistic collaboration of these methods, AgentInfer reduces ineffective token consumption by over 50%, achieving an overall 1.8-2.5 times speedup with preserved accuracy. These results underscore that optimizing for agentic task completion-rather than merely per-token throughput-is the key to building scalable, efficient, and self-improving intelligent systems.

cs.CL↗

arXiv2026

What Matters For Safety Alignment?

This paper presents a comprehensive empirical study on the safety alignment capabilities. We evaluate what matters for safety alignment in LLMs and LRMs to provide essential insights for developing more secure and reliable AI systems. We systematically investigate and compare the influence of six critical intrinsic model characteristics and three external attack techniques. Our large-scale evaluation is conducted using 32 recent, popular LLMs and LRMs across thirteen distinct model families, spanning a parameter scale from 3B to 235B. The assessment leverages five established safety datasets and probes model vulnerabilities with 56 jailbreak techniques and four CoT attack strategies, resulting in 4.6M API calls. Our key empirical findings are fourfold. First, we identify the LRMs GPT-OSS-20B, Qwen3-Next-80B-A3B-Thinking, and GPT-OSS-120B as the top-three safest models, which substantiates the significant advantage of integrated reasoning and self-reflection mechanisms for robust safety alignment. Second, post-training and knowledge distillation may lead to a systematic degradation of safety alignment. We thus argue that safety must be treated as an explicit constraint or a core optimization objective during these stages, not merely subordinated to the pursuit of general capability. Third, we reveal a pronounced vulnerability: employing a CoT attack via a response prefix can elevate the attack success rate by 3.34x on average and from 0.6% to 96.3% for Seed-OSS-36B-Instruct. This critical finding underscores the safety risks inherent in text-completion interfaces and features that allow user-defined response prefixes in LLM services, highlighting an urgent need for architectural and deployment safeguards. Fourth, roleplay, prompt injection, and gradient-based search for adversarial prompts are the predominant methodologies for eliciting unaligned behaviors in modern models.

cs.CL↗

arXiv2026

ReThinker: Scientific Reasoning by Rethinking with Guided Reflection and Confidence Control

Expert-level scientific reasoning remains challenging for large language models, particularly on benchmarks such as Humanity's Last Exam (HLE), where rigid tool pipelines, brittle multi-agent coordination, and inefficient test-time scaling often limit performance. We introduce ReThinker, a confidence-aware agentic framework that orchestrates retrieval, tool use, and multi-agent reasoning through a stage-wise Solver-Critic-Selector architecture. Rather than following a fixed pipeline, ReThinker dynamically allocates computation based on model confidence, enabling adaptive tool invocation, guided multi-dimensional reflection, and robust confidence-weighted selection. To support scalable training without human annotation, we further propose a reverse data synthesis pipeline and an adaptive trajectory recycling strategy that transform successful reasoning traces into high-quality supervision. Experiments on HLE, GAIA, and XBench demonstrate that ReThinker consistently outperforms state-of-the-art foundation models with tools and existing deep research systems, achieving state-of-the-art results on expert-level reasoning tasks.

cs.AI↗

arXiv2026

Why Attention Patterns Exist: A Unifying Temporal Perspective Analysis

Attention patterns play a crucial role in both training and inference of large language models (LLMs). Prior works have identified individual patterns such as retrieval heads, sink heads, and diagonal traces, yet these observations remain fragmented and lack a unifying explanation. To bridge this gap, we introduce \textbf{Temporal Attention Pattern Predictability Analysis (TAPPA), a unifying framework that explains diverse attention patterns by analyzing their underlying mathematical formulations} from a temporally continuous perspective. TAPPA both deepens the understanding of attention behavior and guides inference acceleration approaches. Specifically, TAPPA characterizes attention patterns as predictable patterns with clear regularities and unpredictable patterns that appear effectively random. Our analysis further reveals that this distinction can be explained by the degree of query self-similarity along the temporal dimension. Focusing on the predictable patterns, we further provide a detailed mathematical analysis of three representative cases through the joint effect of queries, keys, and Rotary Positional Embeddings (RoPE). We validate TAPPA by applying its insights to KV cache compression and LLM pruning tasks. Across these tasks, a simple metric motivated by TAPPA consistently improves performance over baseline methods. The code is available at https://github.com/MIRALab-USTC/LLM-TAPPA.

cs.CL↗

arXiv2026

Beyond Speedup -- Utilizing KV Cache for Sampling and Reasoning

KV caches, typically used only to speed up autoregressive decoding, encode contextual information that can be reused for downstream tasks at no extra cost. We propose treating the KV cache as a lightweight representation, eliminating the need to recompute or store full hidden states. Despite being weaker than dedicated embeddings, KV-derived representations are shown to be sufficient for two key applications: \textbf{(i) Chain-of-Embedding}, where they achieve competitive or superior performance on Llama-3.1-8B-Instruct and Qwen2-7B-Instruct; and \textbf{(ii) Fast/Slow Thinking Switching}, where they enable adaptive reasoning on Qwen3-8B and DeepSeek-R1-Distil-Qwen-14B, reducing token generation by up to $5.7\times$ with minimal accuracy loss. Our findings establish KV caches as a free, effective substrate for sampling and reasoning, opening new directions for representation reuse in LLM inference. Code: https://github.com/cmd2001/ICLR2026_KV-Embedding.

cs.CL↗

arXiv2026

High-Rank Structured Modulation for Parameter-Efficient Fine-Tuning

As the number of model parameters increases, parameter-efficient fine-tuning (PEFT) has become the go-to choice for tailoring pre-trained large language models. Low-rank Adaptation (LoRA) uses a low-rank update method to simulate full parameter fine-tuning, which is widely used to reduce resource requirements. However, decreasing the rank encounters challenges with limited representational capacity when compared to full parameter fine-tuning. We present \textbf{SMoA}, a high-rank \textbf{S}tructured \textbf{MO}dulation \textbf{A}dapter that uses fewer trainable parameters while maintaining a higher rank, thereby improving the model's representational capacity and offering improved performance potential. The core idea is to freeze the original pretrained weights and selectively amplify or suppress important features of the original weights across multiple subspaces. The subspace mechanism provides an efficient way to increase the capacity and complexity of a model. We conduct both theoretical analyses and empirical studies on various tasks. Experiment results show that SMoA outperforms LoRA and its variants on 10 tasks, with extensive ablation studies validating its effectiveness.

cs.CL↗

arXiv2025

KVTuner: Sensitivity-Aware Layer-Wise Mixed-Precision KV Cache Quantization for Efficient and Nearly Lossless LLM Inference

KV cache quantization can improve Large Language Models (LLMs) inference throughput and latency in long contexts and large batch-size scenarios while preserving LLMs effectiveness. However, current methods have three unsolved issues: overlooking layer-wise sensitivity to KV cache quantization, high overhead of online fine-grained decision-making, and low flexibility to different LLMs and constraints. Therefore, we theoretically analyze the inherent correlation of layer-wise transformer attention patterns to KV cache quantization errors and study why key cache is generally more important than value cache for quantization error reduction. We further propose a simple yet effective framework KVTuner to adaptively search for the optimal hardware-friendly layer-wise KV quantization precision pairs for coarse-grained KV cache with multi-objective optimization and directly utilize the offline searched configurations during online inference. To reduce the computational cost of offline calibration, we utilize the intra-layer KV precision pair pruning and inter-layer clustering to reduce the search space. Experimental results show that we can achieve nearly lossless 3.25-bit mixed precision KV cache quantization for LLMs like Llama-3.1-8B-Instruct and 4.0-bit for sensitive models like Qwen2.5-7B-Instruct on mathematical reasoning tasks. The maximum inference throughput can be improved by 21.25\% compared with KIVI-KV8 quantization over various context lengths. Our code and searched configurations are available at https://github.com/cmd2001/KVTuner.

cs.LG↗

arXiv2025

Towards Prospective Medical Image Reconstruction via Knowledge-Informed Dynamic Optimal Transport

Medical image reconstruction from measurement data is a vital but challenging inverse problem. Deep learning approaches have achieved promising results, but often requires paired measurement and high-quality images, which is typically simulated through a forward model, i.e., retrospective reconstruction. However, training on simulated pairs commonly leads to performance degradation on real prospective data due to the retrospective-to-prospective gap caused by incomplete imaging knowledge in simulation. To address this challenge, this paper introduces imaging Knowledge-Informed Dynamic Optimal Transport (KIDOT), a novel dynamic optimal transport framework with optimality in the sense of preserving consistency with imaging physics in transport, that conceptualizes reconstruction as finding a dynamic transport path. KIDOT learns from unpaired data by modeling reconstruction as a continuous evolution path from measurements to images, guided by an imaging knowledge-informed cost function and transport equation. This dynamic and knowledge-aware approach enhances robustness and better leverages unpaired data while respecting acquisition physics. Theoretically, we demonstrate that KIDOT naturally generalizes dynamic optimal transport, ensuring its mathematical rationale and solution existence. Extensive experiments on MRI and CT reconstruction demonstrate KIDOT's superior performance.

eess.IV↗

arXiv2025

A Helping (Human) Hand in Kinematic Structure Estimation

Visual uncertainties such as occlusions, lack of texture, and noise present significant challenges in obtaining accurate kinematic models for safe robotic manipulation. We introduce a probabilistic real-time approach that leverages the human hand as a prior to mitigate these uncertainties. By tracking the constrained motion of the human hand during manipulation and explicitly modeling uncertainties in visual observations, our method reliably estimates an object's kinematic model online. We validate our approach on a novel dataset featuring challenging objects that are occluded during manipulation and offer limited articulations for perception. The results demonstrate that by incorporating an appropriate prior and explicitly accounting for uncertainties, our method produces accurate estimates, outperforming two recent baselines by 195% and 140%, respectively. Furthermore, we demonstrate that our approach's estimates are precise enough to allow a robot to manipulate even small objects safely.

cs.RO↗

arXiv2025

AttentionPredictor: Temporal Patterns Matter for KV Cache Compression

With the development of large language models (LLMs), efficient inference through Key-Value (KV) cache compression has attracted considerable attention, especially for long-context generation. To compress the KV cache, recent methods identify critical KV tokens through static modeling of attention scores. However, these methods often struggle to accurately determine critical tokens as they neglect the temporal patterns in attention scores, resulting in a noticeable degradation in LLM performance. To address this challenge, we propose AttentionPredictor, which is the first learning-based method to directly predict attention patterns for KV cache compression and critical token identification. Specifically, AttentionPredictor learns a lightweight, unified convolution model to dynamically capture spatiotemporal patterns and predict the next-token attention scores. An appealing feature of AttentionPredictor is that it accurately predicts the attention score and shares the unified prediction model, which consumes negligible memory, among all transformer layers. Moreover, we propose a cross-token critical cache prefetching framework that hides the token estimation time overhead to accelerate the decoding stage. By retaining most of the attention information, AttentionPredictor achieves 13$\times$ KV cache compression and 5.6$\times$ speedup in a cache offloading scenario with comparable LLM performance, significantly outperforming the state-of-the-arts. The code is available at https://github.com/MIRALab-USTC/LLM-AttentionPredictor.

cs.CL↗

arXiv2025

The direct linearization scheme with the Lamé function: The KP equation and reductions

The paper starts from establishing an elliptic direct linearization (DL) scheme for the Kadomtsev-Petviashvili equation. The scheme consists of an integral equation (involving the Lamé function) and a formula for elliptic soliton solutions, which can be confirmed by checking Lax pair. Based on analysis of real-valuedness of the Weierstrass functions, we are able to construct a Marchenko equation for elliptic solitons. A mechanism to obtain nonsingular real solutions from this elliptic DL scheme is formulated. By utilizing elliptic $N$th roots of unity and reductions, the elliptic DL schemes, Marchenko equations and nonsingular real solutions are studied for the Korteweg-de Vries equation and Boussinesq equation. Illustrations of the obtained solutions show solitons and their interactions on a periodic background.

nlin.SI↗

arXiv2025

Behavioral Fingerprinting of Large Language Models

Current benchmarks for Large Language Models (LLMs) primarily focus on performance metrics, often failing to capture the nuanced behavioral characteristics that differentiate them. This paper introduces a novel ``Behavioral Fingerprinting'' framework designed to move beyond traditional evaluation by creating a multi-faceted profile of a model's intrinsic cognitive and interactive styles. Using a curated \textit{Diagnostic Prompt Suite} and an innovative, automated evaluation pipeline where a powerful LLM acts as an impartial judge, we analyze eighteen models across capability tiers. Our results reveal a critical divergence in the LLM landscape: while core capabilities like abstract and causal reasoning are converging among top models, alignment-related behaviors such as sycophancy and semantic robustness vary dramatically. We further document a cross-model default persona clustering (ISTJ/ESTJ) that likely reflects common alignment incentives. Taken together, this suggests that a model's interactive nature is not an emergent property of its scale or reasoning power, but a direct consequence of specific, and highly variable, developer alignment strategies. Our framework provides a reproducible and scalable methodology for uncovering these deep behavioral differences. Project: https://github.com/JarvisPei/Behavioral-Fingerprinting

cs.CL↗

arXiv2025

Scaling Up, Speeding Up: A Benchmark of Speculative Decoding for Efficient LLM Test-Time Scaling

Test-time scaling has emerged as a powerful paradigm for enhancing the reasoning capabilities of large language models (LLMs) by allocating additional computational resources during inference. However, this paradigm is inherently inefficient due to the generation of redundant and repetitive reasoning traces, leading to significant computational overhead. Speculative decoding offers a promising avenue for mitigating this inefficiency, yet its efficacy in the structured, repetition-rich context of test-time scaling remains largely unexplored. To bridge this gap, we introduce the first comprehensive benchmark designed to evaluate speculative decoding methods for accelerating LLM test-time scaling. Our benchmark provides consistent experimental protocols across representative test-time scaling paradigms (e.g., Best-of-N sampling and multi-round thinking), enabling a fair comparison of three major categories of speculative decoding: model-based, training-based, and n-gram-based methods. Extensive experiments reveal that simple n-gram-based methods effectively capture repetitive patterns, demonstrating unique potential in accelerating test-time scaling. This phenomenon demonstrates the value of integrating n-gram-based methods with model-based or training-based approaches to balance acceleration for both repetitive and diverse reasoning in test-time scaling. We hope this benchmark spurs further research on speculative decoding for test-time scaling, enabling faster and more practical reasoning in LLMs through better handling of repetitive and diverse reasoning paths.

cs.CL↗

arXiv2025

ELF: Efficient Logic Synthesis by Pruning Redundancy in Refactoring

In electronic design automation, logic optimization operators play a crucial role in minimizing the gate count of logic circuits. However, their computation demands are high. Operators such as refactor conventionally form iterative cuts for each node, striving for a more compact representation - a task which often fails 98% on average. Prior research has sought to mitigate computational cost through parallelization. In contrast, our approach leverages a classifier to prune unsuccessful cuts preemptively, thus eliminating unnecessary resynthesis operations. Experiments on the refactor operator using the EPFL benchmark suite and 10 large industrial designs demonstrate that this technique can speedup logic optimization by 3.9x on average compared with the state-of-the-art ABC implementation.

cs.LG↗

arXiv2025

20-GHz bandwidth optical activation function based on a semiconductor laser

Optical neural networks usually execute the linear multiply-accumulate operation in the optical domain, whereas the nonlinear activation function is mostly implemented in the digital or electrical domain. Here we demonstrate a broadband activation function operated in the optical domain. The optical activation function (OAF) is achieved by a semiconductor laser with optical injection, which is operated in the unstable-locked regime above the Hopf bifurcation. The operation bandwidth of the OAF reaches as high as 20 GHz, which is limited by the relaxation oscillation resonance frequency of the semiconductor laser. The OAF mimics a ReLU function for modulation frequencies below 5 GHz, while resembling an ELU function for frequencies above 5 GHz.

physics.optics↗

arXiv2025

Hybrid Near-Far Field 6D Movable Antenna Design Exploiting Directional Sparsity and Deep Learning

Six-dimensional movable antenna (6DMA) has been identified as a new disruptive technology for future wireless systems to support a large number of users with only a few antennas. However, the intricate relationships between the signal carrier wavelength and the transceiver region size lead to inaccuracies in traditional far-field 6DMA channel model, causing discrepancies between the model predictions and the hybrid-field channel characteristics in practical 6DMA systems, where users might be in the far-field region relative to the antennas on the same 6DMA surface, while simultaneously being in the near-field region relative to different 6DMA surfaces. Moreover, due to the high-dimensional channel and the coupled position and rotation constraints, the estimation of the 6DMA channel and the joint design of the 6DMA positions and rotations and the transmit beamforming at the base station (BS) incur extremely high computational complexity. To address these issues, we propose an efficient hybrid-field generalized 6DMA channel model, which accounts for planar-wave propagation within individual 6DMA surfaces and spherical-wave propagation among different 6DMA surfaces. Furthermore, by leveraging directional sparsity, we propose a low-overhead channel estimation algorithm that efficiently constructs a complete channel map for all potential antenna position-rotation pairs while limiting the training overhead incurred by antenna movement. In addition, we propose a low-complexity design leveraging deep reinforcement learning (DRL), which facilitates the joint design of the 6DMA positions, rotations, and beamforming in a unified manner. Numerical results demonstrate that the proposed hybrid-field channel model and channel estimation algorithm outperform existing approaches and that the DRL-enhanced 6DMA system significantly surpasses flexible antenna systems.

cs.IT↗