Searcharxiv⌕ Search

arXiv subjects

Cong Li

Publications and source records attributed to Cong Li.

At least 19 recordsLinked to original sources
arXiv2026

HDA-MoE: Hybrid Parallelism and Dynamic, Adaptive Scheduling for Mixture-of-Experts with 3D Near-Memory Processing

Mixture-of-Experts (MoE) architectures have become a key technique for scaling Large Language Models (LLMs), enabling high model capacity with reduced computational cost. However, this efficiency comes at the expense of increased memory capacity and bandwidth demands. Recent 3D Near-Memory Processing (NMP) architectures, which vertically integrate memory and compute through hybrid bonding, provide high internal bandwidth and energy efficiency, making them attractive for accelerating MoE inference. Nevertheless, the distributed memory and compute organization of NMP systems introduces new challenges for mapping MoE workloads. Existing parallelization strategies, such as Tensor Parallelism (TP) and Expert Parallelism (EP), suffer from either high communication costs or unbalanced computation utilization, leading to inferior efficiency. In addition, the dynamic routing behavior of MoE models further complicates efficient deployment. To address these challenges, we present HDA-MoE, a framework that optimizes MoE execution on NMP architectures through hybrid parallel deployment and runtime scheduling. HDA-MoE integrates an offline hybrid parallel mapping algorithm with an online dynamic and adaptive scheduling mechanism to reduce communication overhead while improving computation utilization. Experimental results show that HDA-MoE achieves a speedup of 1.1x--3.4x over TP, 1.1x--1.5x over EP, 1.1x--3.7x over the Hybrid TP-EP compute-balanced baseline, and 1.1x--1.3x over HD-MoE. Source code is available at https://github.com/PKU-SEC-Lab/HDA-MoE-TCAD26.

cs.AR↗
arXiv2026

Distributed Secure Learning Control for Large-scale Multirobots under Stealthy Actuator Attacks

Distributed learning control for multirobot systems (MRS) offers significant flexibility in presence of uncertainties but lacks provable performance guarantees. A promising direction involves integrating reinforcement learning (RL) into distributed model predictive control (DMPC), leveraging the strengths of RL in nonlinear policy design and the receding-horizon replanning capabilities of DMPC. However, ensuring secure control within such a learning framework under malicious cyber attacks, particularly stealthy ones, remains a critical challenge, because the distributed policies generation depends on information exchange among neighbors, where compromised agents can rapidly influence the behavior of others through the communication network. This article proposes a distributed secure learning control (DSLC) framework for large-scale MRS under malicious, stealthy actuator attacks. Our framework offers two key features: (i) a unified approach that enables secure learning control across various coordination scenarios and (ii) a game-theoretic distributed learning-based predictive control strategy that learns how to balance the attacker and defender through a differential-game based DMPC framework. Specifically, DSLC employs a distributed attacker-actor-critic architecture to learn the optimal defense and attack policies online within each prediction interval. Unlike numerical optimization-based controllers that calculate open-loop control sequences, our method simultaneously generates adversarial attack policies and corresponding defense policies in analytical closed-loop form. The defense policies could be directly generalized to MRS with varying scales and diverse actuator attack probabilities. The effectiveness and scalability of DSLC are validated through comprehensive simulations and real-world experiments in multiple wheeled robots via various control tasks.

cs.RO↗
arXiv2026

NSFlow: End-to-End Differentiable Neuro-Symbolic Optical Flow for Visual Odometry

Sparse optical flow provides stable inter-frame correspondence, playing a key role in Visual Odometry (VO) and Visual-Inertial Odometry (VIO). Classical optimization-based methods, such as Lucas-Kanade (LK), perform well under small displacements but are sensitive to large motions and illumination changes. Modern regression-based learning methods, while more robust in complex scenes, are often computationally heavy and lack explicit geometric consistency, making them less suitable for efficient VO/VIO front-ends. To bridge this gap, we propose a hybrid neuro-symbolic framework that combines the strengths of both paradigms. Our method uses a Convolutional Neural Network (CNN) to extract robust feature representations, which is fed into a differentiable LK optimizer to estimate optical flow in an end-to-end trainable manner. Through implicit differentiation, gradients are propagated across the iterative solver, enabling joint optimization of feature extraction and flow estimation. The resulting system integrates seamlessly into existing VO/VIO pipelines and runs in real-time on embedded platforms. Experiments show that our method outperforms conventional optimization-based flow in challenging conditions such as dynamic lighting and low texture, while also achieving higher accuracy and lower latency than purely regression-based alternatives. When deployed in a VIO system, our method demonstrates significant performance improvement, achieving an average error reduction of 42\% on challenging datasets while enhancing tracking stability. The code is publicly available.

cs.CV↗
arXiv2026

Resolving Local Structure of Distribution Functions in Collider Measurements

Collider measurements may blur narrow features in underlying distributions before fitting, since a data point samples a finite region. We show that observables linear in a selected target distribution, after externally constrained inputs are absorbed into a nonnegative finite response, have a calculable distribution-space locality set by hard dynamics and kinematics. The baseline-normalized $F_{\rm dir}(X)$ yields the local value at zero width and a controlled average otherwise, with covariance setting the leading bias. This guides the choice of processes and observables that retain unexpected fine structure. An EIC photon-radiation observable realizes the exact zero-width limit.

hep-ph↗
arXiv2026

Dimensionality-induced critical phase transition in stochastic Lotka-Volterra equation: From statistical averaging to systemic tipping point

By analyzing a stochastic Lotka-Volterra (LV) equation, we show that the underlying logic behind the diversity-stability debate in ecology can be framed as a dimensionality-induced critical phase transition, that is, a transition separating the regime dominated by statistical averaging and a systemic tipping point. This phase transition framework unifies the opposing ecological predictions in the diversity-stability debate and sets an intrinsic diversity ceiling for stochastic ecological communities.

q-bio.PE↗
arXiv2026

Multi-Image Visual Token Pruning in Large Visual Language Models

With the growing demand for processing multiple image sequences in real-world applications, various visual token pruning methods have emerged to mitigate the computational and context length constraints faced by Large Vision Language Models (LVLMs). However, most existing pruning approaches rely on static strategies that struggle to adapt across different architectural LVLMs and multi-image scenarios, and are additionally constrained by their dependence on attention computations that are incompatible with efficient techniques like FlashAttention. To address these limitations, we propose a training-free, Adaptive Visual Token Pruning (AVTP) framework, applicable to diverse LVLM architectures. We strategically determine pruning layers based on empirical analysis of visual attention distributions across various LVLMs, and implement adaptive pruning ratios in multi-image contexts where images of higher importance retain proportionally more tokens. We conduct extensive experiments across different LVLMs to demonstrate the effectiveness and robustness of AVTP. Specifically, Qwen3VL-8B achieves 2 times inference speedup while maintaining 96.1\% of its original accuracy on multiple multi-image benchmarks, InternVL3.5-8B retains 94.1\% accuracy, and LLaVA-OV-7B even exceeds its original baseline performance. Our code is available at \href{https://github.com/zry13/AVTP}{this link}.

cs.CV↗
arXiv2026

AS-FedBridge: Pseudo-Spike Bridge Distillation for Heterogeneous ANN-SNN Federated Learning

Federated learning enables collaborative model training across distributed edge devices while strictly preserving data privacy. To facilitate practical deployment on resource-constrained edge devices, Spiking Neural Networks (SNNs) have emerged as a promising alternative to traditional Artificial Neural Networks (ANNs) due to their sparse computing mechanisms and high energy efficiency. However, jointly training ANNs and SNNs exposes a challenge of representational misalignment, which is intrinsically caused by differences in information representation, specifically the semantic gap between continuous real-valued activations in ANNs and discrete spatio-temporal spikes in SNNs. To overcome this barrier, we propose AS-FedBridge, a novel federated learning framework tailored for mixed ANN-SNN clients. AS-FedBridge features a lightweight Bridge equipped with a Pseudo-Spike Interface, which effectively projects continuous signals into a spike-compatible space to facilitate ANN-SNN alignment. Given the absence of existing mixed ANN-SNN federated frameworks, we establish a comprehensive benchmark to evaluate against multiple advanced heterogeneous FL methods. Our empirical analysis demonstrates a positive correlation between the degree of ANN-SNN alignment and the collaborative FL performance. Across four datasets, AS-FedBridge consistently demonstrates advanced accuracy while mitigating extreme scale, architecture, and client heterogeneity challenge. Furthermore, our framework enables a highly controllable trade-off between model performance and resource efficiency. AS-FedBridge accomplishes these robust performance gains while introducing only marginal computational overhead, establishing a robust and practical foundation for mixed ANN-SNN federated learning systems.

cs.LG↗
arXiv2026

COMEX: A Composition-Grounded Benchmark and Learning Framework for Explainable Aesthetic Image Cropping

Explainable aesthetic image cropping requires not only localizing a visually pleasing crop but also explaining why it is preferred. Existing crop-and-explain methods largely treat explanation as post-hoc text generation and overlook composition, a key aesthetic factor that links crop decisions with interpretable reasoning. In this paper, we reformulate explainable aesthetic image cropping as a structured crop-composition-explanation problem. To support this setting, we introduce COMEX, a new benchmark built through image expansion and an IO-reversal pipeline. COMEX contains 33,161 quadruples, each consisting of an expanded image, a crop box, a composition category, and a composition-grounded explanation, enabling joint learning of crop localization, composition understanding, and explanation generation. We further propose a two-stage SFT+GRPO framework, where supervised fine-tuning establishes the structured output protocol and basic cropping ability, and GRPO further improves crop quality, composition prediction, and explanation faithfulness. We benchmark 15 large vision-language models and existing cropping methods on COMEX, establishing a comprehensive testbed for composition-grounded explainable aesthetic cropping. Experiments on both COMEX and prior benchmarks demonstrate the effectiveness and transferability of our framework, with strong performance across evaluation metrics.

cs.CV↗
arXiv2026

TAPO: Transition-Aware Policy Optimization for LLM Agents

Recently, Reinforcement Learning (RL) has emerged as a crucial paradigm for the post-training of Large Language Model (LLM) agents. However, existing methods predominantly rely on sparse task rewards for policy optimization, failing to fully exploit another class of inherently dense supervisory signals naturally present during online interaction: environmental feedback following action execution. Recent theoretical studies suggest that generalization in multi-step, goal-oriented tasks hinges on predictive knowledge of environmental consequences. Inspired by this, we propose TAPO: Transition-Aware Policy Optimization for LLM Agents, a unified training framework that alternates between policy optimization and transition supervision. Beyond standard RL updates, TAPO repurposes rollout data to apply action-conditioned next-observation prediction supervision on a shared backbone model. This approach enhances the model's sensitivity to environmental transition dynamics and action consequences while concurrently optimizing the policy. It serves as a computationally lightweight, plug-and-play enhancement module for existing agent RL algorithms, requiring no additional expert data, extra sampling costs, or inference-time overhead. We conduct systematic experiments on WebShop and ALFWorld, integrating foundation models of various scales with different policy optimization algorithms. Empirical results demonstrate that TAPO consistently improves task performance over pure policy optimization baselines.

cs.LG↗
arXiv2026

CODA: Algorithm-Hardware Co-design for Edge Video Diffusion via NMP-Enabled Compute-Cache Operator Disaggregation

Deploying Video Diffusion Models (VDMs) on edge devices is appealing for localized and privacy-preserving generation, but their iterative Transformer-based denoising remains too slow for practical local inference. Cross-Timestep Caching (CTC) has emerged as a promising direction for reducing redundant computation, reusing activations across adjacent denoising steps rather than modifying model weights, while largely preserving generation fidelity. However, on memory-constrained edge GPUs, CTC requires a massive cache footprint that quickly exceeds on-device VRAM and forces the cache into host memory. More fundamentally, cache operators remain tightly interleaved and chain-dependent with native compute operators, so naive near-memory offloading still incurs repeated PCIe exchanges for residual and fusion computations, turning cache reuse into a communication- and serialization-bound execution flow. We therefore propose CODA, an algorithm-hardware co-designed architecture centered on Compute-Cache Operator Disaggregation. CODA separates dense compute paths and memory-bound cache paths across the xPU and a lightweight DIMM-side near-memory engine, reorganizes fragmented cache activity into hardware-friendly coalesced segments, and exploits Classifier-Free Guidance (CFG) branch independence to overlap xPU compute with cache-side execution. Experiments show that CODA achieves up to 1.80x end-to-end speedup and 1.74x higher energy efficiency, while preserving competitive generation quality compared with a state-of-the-art caching algorithm.

cs.AR↗
arXiv2026

Model-Independent and Data-Driven Extraction of the Photon Distribution Function at the Electron--Ion Collider

In this work, we investigate a cross-section-ratio-based inversion method for extracting distribution functions at the Electron--Ion Collider (EIC). Starting from the general factorization formula $d\sigma = H \otimes f$, we show that, for an appropriate bremsstrahlung process and suitably chosen differential observables, the convolution structure can be reduced to a multiplicative form. On this basis, we define the ratio $R = \frac{d\sigma_{\mathrm{exp}}}{d\sigma_{\mathrm{hard}}},$ where $d\sigma_{\mathrm{exp}}$ is the experimentally measured differential cross section and $d\sigma_{\mathrm{hard}}$ is a perturbatively calculable theoretical input. We then establish the relation between $R$ and the target distribution function $f$. Finally, we investigate how the theoretical input $d\sigma_{\mathrm{hard}}$ should be constructed when soft-photon radiation, finite-bin-width effects, and experimental acceptance are taken into account.

hep-ph↗
arXiv2026

A neuromorphic vision system for open-world visual intelligence

Time-efficient and robust visual intelligence remains a critical challenge in unstructured open-world environments, yet current approaches often rely on computationally intensive neural architectures or task-specific sensors with limited versatility. Inspired by biological vision and information bottleneck theory, we report a neuromorphic vision system that performs task-oriented visual intelligence through an information distillation strategy (named as task traction mechanism) implemented on hardware. The system integrates a polarization-sensitive imager with a resistive random-access memory (RRAM) array to progressively distill task-relevant information via light field selection, region of interest extraction, and target anticipation. The neuromorphic vision system conducts visual tasks within an execution time of 193 {\mu}s. Evaluation across eight challenging open-world scenarios shows accuracy improvements of 25.54%, 37.73%, and 36.10% for object tracking, object segmentation, and trajectory prediction, respectively, together with an average 30.6-fold reduction in latency relative to state-of-the-art solutions.

eess.IV↗
arXiv2026

Semi-analytical results for $e^+e^-\to J/\psi + X_{{\rm non\,}c\bar{c}}$ up to $\mathcal{O}(\alpha_s v^2)$ at B factories

Within the NRQCD factorization framework, we investigate the color-singlet contribution to $e^+e^- \to J/\psi + X_{{\rm non\,}c\bar{c}}$ at B factories, computing the $\mathcal{O}(\alpha_s)$, $\mathcal{O}(v^2)$, and $\mathcal{O}(\alpha_s v^2)$ corrections to both the unpolarized cross section and the $J/\psi$ angular distribution. The $\mathcal{O}(\alpha_s v^2)$ correction is obtained for the first time, and the validity of NRQCD factorization at this order is explicitly verified. Using the differential equation method, the short-distance coefficients are obtained as asymptotic expansions in $r = m_c/\sqrt{s}$ up to $r^{40}$, which reproduce exact results with high precision at B factory energies, achieving relative errors around $10^{-14}$ for the cross section and around $10^{-7}$ for the angular distribution. Notably, with the same input parameters, our $\mathcal{O}(\alpha_s)$ and $\mathcal{O}(v^2)$ corrections are consistent with those reported in the literature. Phenomenologically, the $\mathcal{O}(\alpha_s)$ correction (with $\mu_R=\sqrt{s}/2$) reaches about $50\%$ of the leading-order cross section, while the $\mathcal{O}(v^2)$ and $\mathcal{O}(\alpha_s v^2)$ corrections are accidentally small. After including feeddown contributions from $\psi(2S)$, the predicted cross section $0.523_{-0.197}^{+0.285}$ pb agrees with the {\tt Belle} measurement within uncertainties. However, the predicted angular distribution parameter $0.120_{-0.036}^{+0.041}$ deviates from the experimental value $5.71\pm 2.51$ by more than $2\sigma$, calling for further experimental and theoretical investigations.

hep-ph↗
arXiv2026

Reinforcement learning in linear embedding space unlocks generalizable control across soft robot configurations

Soft-bodied organisms such as octopuses and elephant trunks exhibit remarkable morphological adaptability, dynamically reconfiguring body shape and stiffness, and flexibly adjusting their control strategies to enable versatile behaviors. Inspired by these biological systems, various soft robots have emerged in recent decades, featuring diverse materials, stiffnesses, and morphologies tailored to specific tasks. Despite substantial advances in the materials and structural designs of soft robots, developing a generalizable control framework capable of rapid adaptation across diverse configurations remains a long-standing challenge. Existing controllers are limited to fixed configurations, demanding laborious configuration-specific remodelling and policy redesign for new configurations. Here, we introduce a generalizable control system that enables rapid adaptation across diverse soft robot configurations via reinforcement learning in a shared linear Koopman embedding space. By encoding robot dynamics into this embedding space, our method decouples control policies from specific morphologies, allowing real-time, model-free policy adaptation across diverse configurations without retraining from scratch. We validate our system across 33 distinct robot configurations. Our system achieves a 75 times reduction in transfer samples across configurations, while sustaining robust performance under high-speed motion, heavy payloads, and multiactuator faults, and achieving real-world skills previously unattainable in soft robotics. This work establishes a unified and adaptable control paradigm for diverse soft robot configurations, bridging mechanical reconfigurability with control flexibility, and may offer broader insights for generalizable control in complex physical systems.

cs.RO↗
arXiv2026

Your UnEmbedding Matrix is Secretly a Feature Lens for Text Embeddings

Large language models exhibit impressive zero-shot capabilities across a wide range of downstream tasks. However, they struggle to function as off-the-shelf embedding models, leading to suboptimal performance on massive text embedding benchmarks. In this paper, we identify a potential cause underlying this deficiency. Our motivation stems from an unexpected observation: text embeddings tend to align with frequent but uninformative tokens when projected onto the vocabulary space. We argue that this excessive expression of high-frequency tokens suppresses the model's ability to capture nuanced semantics. To address this, we introduce EmbedFilter, a simple linear transformation designed to refine text embeddings derived from LLMs directly. Specifically, we uncover that the unembedding matrix within LLMs encodes a latent space that is actively writing these frequent tokens into embedding space. By filtering out this subspace, EmbedFilter suppress the influence of high-frequency tokens, thereby enhancing semantic representations. As a compelling byproduct, this enables an inherent dimensionality reduction, lowering index storage and speedup retrieval while fully preserving the refined embedding quality. Our experiments across multiple LLM backbones demonstrate that LLMs equipped with EmbedFilter achieve superior zero-shot downstream performance even with significantly reduced embedding dimensions. We hope our findings provide deeper insights into the mechanisms of LLM-based representations and inspire more principled designs to improve text embeddings training. Our code is available at https://github.com/CentreChen/EmbFilter.

cs.CL↗
arXiv2026

Hydrogen-induced lattice cohesion weakening favors atomic displacement

Atomic displacement -- the fundamental process underlying diverse deformation and damage phenomena in metals, from irradiation defect production to stress-driven dislocation motion -- is governed by interatomic cohesion strength. Here, lattice-dissolved hydrogen (LDH) occurring in metals under direct hydrogen exposure is identified to effectively weaken lattice cohesion, and thereby facilitating atomic displacement and dislocation movement upon plastic deformation in sub-threshold stress regime. This atomic-scale insight provides a physically transparent mechanism for hydrogen-enhanced localized plasticity implicated in hydrogen embrittlement. We quantitatively verify the hydrogen-induced lattice cohesion weakening effect on metal surfaces exposed to low-energy hydrogen plasma, where massive defects are generated despite the absence of sufficient ion momentum for direct displacement damage. By unprecedentedly quantifying the cohesion-weakening effect of LDH independently from defect-trapped H, we establish a new paradigm to understand hydrogen embrittlement.

cond-mat.mtrl-sci↗
arXiv2026

Stress-triggered atomic explosion of trapped hydrogen initiates crack nucleation

Hydrogen embrittlement (HE) has persisted for more than a century as one of the most intractable problems in materials science. The prevailing view1 that diffusive H governs embrittlement has fostered the widespread assumption that H trapping at crystal defects mitigates HE. Here we overturn this conventional paradigm. Using plasma/ion irradiation of tungsten, we decouple -- for the first time -- H-induced crack nucleation from subsequent cavity propagation, and reveal nucleation as a two-stage mechanochemical fracture instability enabled by trapped H in the absence of diffusive H. In the first stage, H accumulation to a critical occupancy at dislocation cores acts as a chemical fuse, collapsing the local cohesive strength to a threshold at which infinitesimal external loads can trigger atomic decohesion. This bond rupture instantaneously enables the second stage: confined recombination of atomic hydrogen into molecular form. The abrupt release of chemical energy within an atomically restricted volume generates a transient inflation pressure that drives a dynamic, brittle jump to an internal macroscopic cavity. By separating mechanical decohesion triggering from energetic crack driving, our results provide a deterministic framework for the onset of H-induced crack nucleation under low-stress conditions. Furthermore, we place experimentally the classical H-enhanced decohesion model on an atomistic foundation and elevate it from phenomenology to prediction. Finally, by shifting the focus from experimentally elusive diffusive H to directly measurable trapped H, this work reframes HE as a deterministic, quantifiable instability, establishing a new paradigm for understanding and mitigating H-induced failure in high-strength metals.

cond-mat.mtrl-sci↗
arXiv2026

Semi-analytical two-loop QCD corrections to $e^+e^-\to J/\psi+\chi_{cJ}$ at B factories

In this work, we compute the next-to-next-to-leading-order (NNLO) QCD corrections to the process $e^+e^-\to J/\psi+\chi_{cJ}$ at B factories within the NRQCD factorization framework. The helicity amplitudes are obtained via asymptotic expansions around $r=0$ and $r=1$, with $r=16m_c^2/s$. Our asymptotic expressions reproduce the exact numerical results with high accuracy across the entire range $0\le r \le 1$, achieving a relative error below $10^{-5}$, which is sufficient for phenomenological applications. Notably, the large logarithmic terms are obtained analytically. We compute the unpolarized cross sections. The $\mathcal{O}(\alpha_s)$ correction is found to be large, while the $\mathcal{O}(\alpha_s^2)$ correction for $\chi_{c0}$ production amounts to $33\%$ of the leading-order (LO) cross section, significantly reducing the scale uncertainties. For $\chi_{c1}$, the $\mathcal{O}(\alpha_s)$ and $\mathcal{O}(\alpha_s^2)$ corrections correspond to $35\%$ and $-15\%$, respectively. For $\chi_{c2}$, the corresponding corrections are $25\%$ and $-38\%$. The large cancellation between the corrections for $\chi_{c2}$ brings the NNLO cross section close to the LO prediction. Our prediction for $\chi_{c0}$ is consistent with the {\tt Belle} measurement and agrees with the {\tt BaBar} data within $2\sigma$. We also predict the angular distribution parameters $\alpha^J_\theta$, which are independent of nonperturbative inputs. A sharp discrepancy between the theory and the {\tt Belle} measurement is observed for $\alpha^0_\theta$, calling for further experimental and theoretical investigations. Moreover, future measurements of the angular distribution parameters for $\chi_{c1}$ and $\chi_{c2}$ will provide important tests of the theoretical framework.

hep-ph↗