Search arXiv⌕ Search

arXiv subjects

Junyu Lin

Publications and source records attributed to Junyu Lin.

At least 19 recordsLinked to original sources
arXiv2026

PGuard: A Native, Streaming, and Plug-in Guardrail for Multimodal Large Language Models

As streaming generation becomes the dominant serving paradigm for both text-only and multimodal large language models, ensuring safety throughout the generation process has become a critical challenge. We introduce PGuard, a native, streaming, and plug-in guardrail that transforms safety monitoring from an external serving dependency into an integrated capability of the backbone model. PGuard directly leverages the backbone's intermediate representations for token-level safety detection, without altering its original inference pathway or generation behavior. Specifically, dynamic layer gating selectively aggregates residual-stream representations across layers to extract safety-relevant signals during both prefill and autoregressive decoding. To balance detection accuracy and computational efficiency, PGuard supports multiple representation-modeling architectures, including a lightweight stateless variant and sequence-aware variants built on multi-scale causal EMA-attention and GRU-based recurrent modules. Its design also accounts for the execution characteristics of LLM serving frameworks, such as SGLang, optimizing computation and data access to minimize additional runtime and synchronization overhead. As a result, PGuard incurs only 0.82-2.26% end-to-end overhead on the hundred-billion-parameter backbone Qwen3.5-122B-A10B served by SGLang. Extensive evaluations across more than 30 benchmarks spanning text and multimodal safety, jailbreaks, cyber and CBRN risks, multilingual scenarios, and real-world traffic demonstrate state-of-the-art performance among native guardrails. On text queries with Qwen3.5-9B, PGuard achieves an average F1 score of 0.836, approaching dedicated external guardrails, while supporting deployment-time configuration of category-specific thresholds and enforcement actions.

cs.CL↗
arXiv2026

Is High Local Accuracy Enough? Geometric Consensus with Support-Adaptive Statistical Personalization for Federated Learning

Real-world deployment of personalized federated learning (PFL) requires adapting to heterogeneous clients while preserving recognition performance beyond local distributions. However, high local accuracy can conceal failures on locally rare or absent classes. Effective personalization should therefore strengthen global recognition while raising the personalization floor. We propose an evaluation protocol that complements local accuracy with global, rare-class, and unseen-class evaluations to reveal generalization obscured by local scores, and assesses personalization floors among the worst-performing clients to expose weaknesses hidden by aggregate performance. To meet these deployment challenges, we propose FedGRC, a unified PFL framework integrating generalization and personalization through (i) Consensus-Guided Representation Learning (CGRL), which learns discriminative representations through shared directional and magnitude guidance; (ii) Multilayer Metric Synthesis (MMS), which constructs a shared covariance metric from these multilayer representations for statistical classification; and (iii) Support-Adaptive Shrinkage Personalization (SASP), which adapts class means within this metric by shrinking local estimates toward shared means according to local sample support, enabling adaptation while retaining shared knowledge. Extensive experiments, including benchmarking against 21 methods across 5 heterogeneity settings on 3 datasets and comprehensive ablation studies, validate the effectiveness of FedGRC. On CIFAR-10 and CIFAR-100, FedGRC surpasses state-of-the-art (SOTA) baselines in global recognition and personalization floors by at least 2.17% and 2.13%, respectively. On the real-world heterogeneous FedRS dataset, it outperforms the respective best baselines by 2.15%, 3.30%, and 4.61% in local adaptation, global recognition, and the personalization floor.

cs.LG↗
arXiv2026

Aging-Aware Online Distributed Scheduling for Lifecycle Carbon Reduction in Geo-Distributed Data Centers

The rapid proliferation of data centers (DCs), driven by cloud computing and artificial intelligence (AI), has led to massive energy demand and carbon emissions, posing significant sustainability challenges. Carbon-aware optimization in geographically distributed data centers has been widely studied. Most existing approaches mainly focus on operational carbon emissions from server usage. However, existing literature often ignores workload-induced thermal stress, which accelerates nonlinear hardware degradation. This leads to more frequent server replacements and ultimately increases embodied carbon emissions. To address these limitations, we propose a comprehensive carbon life-cycle modeling framework for distributed data centers. Apart from operational carbon emissions, this work combines workload scheduling with a utilization-dependent exponential aging model to evaluate long-term carbon costs from server degradation. In order to solve the proposed optimization model in an online and privacy-preserving manner, an enhanced Lyapunov framework with time-varying queue shifting (TVQS) is first introduced to handle system uncertainties. Then, a zero-sum perturbation-based alternating direction method of multipliers (ZSP-ADMM) framework is developed to enable distributed coordination across geographically separated data centers while protecting locally exchanged workload information. Simulation results demonstrate that the proposed approach achieves up to 13.0% lower carbon emissions and 12.6% lower operational costs compared with benchmarks.

eess.SY↗
arXiv2026

SkillNet: Create, Evaluate, and Connect AI Skills

Current AI agents can flexibly invoke tools and execute complex tasks, yet their long-term advancement is hindered by the lack of systematic accumulation and transfer of skills. Without a unified mechanism for skill consolidation, agents frequently ``reinvent the wheel'', rediscovering solutions in isolated contexts without leveraging prior strategies. To address this challenge, we introduce SkillNet, an open infrastructure for creating, evaluating, and organizing AI skills at scale. SkillNet structures skills within a unified ontology that supports creating skills from heterogeneous sources, establishing rich relational connections, and performing multi-dimensional evaluation across Safety, Completeness, Executability, Maintainability, and Cost-awareness. Our infrastructure integrates a repository of over 600,000 skills, an interactive platform, and a versatile Python toolkit. Experiments on ALFWorld, WebShop, and ScienceWorld show 40% higher average rewards and 30% fewer execution steps across multiple backbone models. Furthermore, SkillNet-Gym benchmarks skill retrieval, utilization, and composition, while SkillNet-Fabric enables task-specific skill routing through lightweight Wikis. By formalizing skills as evolving, composable assets, SkillNet provides a robust foundation for agents to move from transient experience to durable mastery.

cs.AI↗
arXiv2026

YuFeng-XGuard: A Reasoning-Centric, Interpretable, and Flexible Guardrail Model for Large Language Models

As large language models (LLMs) are increasingly deployed in real-world applications, safety guardrails are required to go beyond coarse-grained filtering and support fine-grained, interpretable, and adaptable risk assessment. However, existing solutions often rely on rapid classification schemes or post-hoc rules, resulting in limited transparency, inflexible policies, or prohibitive inference costs. To this end, we present YuFeng-XGuard, a reasoning-centric guardrail model family designed to perform multi-dimensional risk perception for LLM interactions. Instead of producing opaque binary judgments, YuFeng-XGuard generates structured risk predictions, including explicit risk categories and configurable confidence scores, accompanied by natural language explanations that expose the underlying reasoning process. This formulation enables safety decisions that are both actionable and interpretable. To balance decision latency and explanatory depth, we adopt a tiered inference paradigm that performs an initial risk decision based on the first decoded token, while preserving ondemand explanatory reasoning when required. In addition, we introduce a dynamic policy mechanism that decouples risk perception from policy enforcement, allowing safety policies to be adjusted without model retraining. Extensive experiments on a diverse set of public safety benchmarks demonstrate that YuFeng-XGuard achieves stateof-the-art performance while maintaining strong efficiency-efficacy trade-offs. We release YuFeng-XGuard as an open model family, including both a full-capacity variant and a lightweight version, to support a wide range of deployment scenarios.

cs.CL↗
arXiv2026

Fermi gas of polar molecules in the Pauli-blocked regime

Quantum gases of polar molecules have recently emerged as a powerful platform for exploring exotic many-body dynamics and correlated quantum behavior. To achieve the full potential of this platform, the production of deeply degenerate quantum gases of molecules in arbitrary confinement geometries is necessary. Here, we successfully evaporate fermionic KRb molecules in both 3D and quasi-2D geometries to well below their Fermi temperatures utilizing dipolar collisions. As we evaporate deeper into degeneracy in both geometries, we enter the Pauli-blocked regime with polar molecules, which we independently confirm for the first time by measuring the Pauli suppression of elastic collisions. Moreover, the Pauli suppression of collisions contributes to the limitation of our final molecular temperature to about 25% of the Fermi temperature in both geometries, particularly limiting quasi-2D evaporation where the Pauli blockade drastically reduces an otherwise large elastic to inelastic scattering ratio. This work demonstrates the production of degenerate Fermi gases of polar molecules both in a 3D harmonic trap and in mono- and bi-layer 2D configurations. Further, our work explores the fundamental limits on evaporation of molecular Fermi gases set by the Pauli-exclusion principle, which could be overcome in the future by introducing distinguishable scattering partners.

cond-mat.quant-gas↗
arXiv2025

Observation of generalized t-J spin dynamics with tunable dipolar interactions

Long-range and anisotropic dipolar interactions profoundly modify the dynamics of particles hopping in a periodic lattice potential. We report the realization of a generalized t-J model with dipolar interactions using a system of ultracold fermionic molecules with spin encoded in the two lowest rotational states. We independently tuned the dipolar Ising and spin-exchange couplings and the molecular motion and studied their interplay on coherent spin dynamics. Using Ramsey spectroscopy, we observed and modeled interaction-driven contrast decay that depends strongly both on the strength of the anisotropy between Ising and spin-exchange couplings and on motion. This study paves the way for future exploration of kinetic spin dynamics and quantum magnetism with highly tunable molecular platforms in regimes that are challenging for existing numerical and analytical methods.

cond-mat.quant-gas↗
arXiv2024

Two-axis twisting using Floquet-engineered XYZ spin models with polar molecules

Polar molecules confined in an optical lattice are a versatile platform to explore spin-motion dynamics based on strong, long-range dipolar interactions. The precise tunability of Ising and spin-exchange interactions with both microwave and dc electric fields makes the molecular system particularly suitable for engineering complex many-body dynamics. Here, we used Floquet engineering to realize interesting quantum many-body systems of polar molecules. Using a spin encoded in the two lowest rotational states of ultracold KRb molecules, we mutually validated XXZ spin models tuned by a Floquet microwave pulse sequence against those tuned by a dc electric field through observations of Ramsey contrast dynamics, setting the stage for the realization of Hamiltonians inaccessible with static fields. In particular, we observed two-axis twisting mean-field dynamics, generated by a Floquet-engineered XYZ model using itinerant molecules in 2D layers. In the future, Floquet-engineered Hamiltonians could generate entangled states for molecule-based precision measurement or could take advantage of the rich molecular structure for quantum simulation of multi-level systems.

cond-mat.quant-gas↗
arXiv2023

Revisiting Multi-modal 3D Semantic Segmentation in Real-world Autonomous Driving

LiDAR and camera are two critical sensors for multi-modal 3D semantic segmentation and are supposed to be fused efficiently and robustly to promise safety in various real-world scenarios. However, existing multi-modal methods face two key challenges: 1) difficulty with efficient deployment and real-time execution; and 2) drastic performance degradation under weak calibration between LiDAR and cameras. To address these challenges, we propose CPGNet-LCF, a new multi-modal fusion framework extending the LiDAR-only CPGNet. CPGNet-LCF solves the first challenge by inheriting the easy deployment and real-time capabilities of CPGNet. For the second challenge, we introduce a novel weak calibration knowledge distillation strategy during training to improve the robustness against the weak calibration. CPGNet-LCF achieves state-of-the-art performance on the nuScenes and SemanticKITTI benchmarks. Remarkably, it can be easily deployed to run in 20ms per frame on a single Tesla V100 GPU using TensorRT TF16 mode. Furthermore, we benchmark performance over four weak calibration levels, demonstrating the robustness of our proposed approach.

cs.CV↗
arXiv2023

Microwave shielding of bosonic NaRb molecules

Recent years have witnessed tremendous progresses in creating and manipulating ground-state ultracold polar molecules. However, the two-body loss regardless of the chemical reactivities is still a hurdle for many future explorations. Here, we investigate the loss suppression of non-reactive bosonic $^{23}$Na$^{87}$Rb molecules with a circular polarized microwave blue-detuned to the rotational transition. We achieve suppression of the loss by two orders of magnitude with the lowest two-body loss rate coefficient reduced to $3\times10^{-12}~\rm{cm^3/s}$. Meanwhile, the elastic collision rate coefficient is increased to the $10^{-8}~\rm{cm^3/s}$ level. The large good-to-bad collision ratio has allowed us to carry out evaporative cooling of $^{23}$Na$^{87}$Rb with an efficiency of 1.7(2), increasing the phase-space density by a factor of 10. With further improvements, this technique holds great promises for creating a Bose-Einstein condensate of ultracold polar molecules.

physics.atom-ph↗
arXiv2022

D^2ETR: Decoder-Only DETR with Computationally Efficient Cross-Scale Attention

DETR is the first fully end-to-end detector that predicts a final set of predictions without post-processing. However, it suffers from problems such as low performance and slow convergence. A series of works aim to tackle these issues in different ways, but the computational cost is yet expensive due to the sophisticated encoder-decoder architecture. To alleviate this issue, we propose a decoder-only detector called D^2ETR. In the absence of encoder, the decoder directly attends to the fine-fused feature maps generated by the Transformer backbone with a novel computationally efficient cross-scale attention module. D^2ETR demonstrates low computational complexity and high detection accuracy in evaluations on the COCO benchmark, outperforming DETR and its variants.

cs.CV↗
arXiv2021

Seconds-scale coherence on nuclear spin transitions of ultracold polar molecules in 3D optical lattices

Ultracold polar molecules (UPMs) are emerging as a novel and powerful platform for fundamental applications in quantum science. Here, we report characterization of the coherence between nuclear spin levels of ultracold ground-state sodium-rubidium molecules loaded into a 3D optical lattice with a nearly photon scattering limited trapping lifetime of 9(1) seconds. After identifying and compensating the main sources of decoherence, we achieve a maximum nuclear spin coherence time of $T_2^* = 3.3(6)$~s with two-photon Ramsey spectroscopy. Furthermore, based on the understanding of the main factor limiting the coherence of the two-photon Rabi transition, we obtain a Rabi lineshape with linewidth below 0.8 Hz. The simultaneous realization of long lifetime and coherence time, and ultra-high spectroscopic resolution in our system unveils the great potentials of UPMs in quantum simulation, computation, and metrology.

physics.atom-ph↗
arXiv2021

Characterization of the lowest excited-state ro-vibrational level of $^{23}$Na$^{87}$Rb

Starting from an ultracold sample of ground-state $^{23}$Na$^{87}$Rb molecules, we investigate the lowest ro-vibrational level of the $b^3Π$ state with high resolution laser spectroscopy. This electronic spin-forbidden $X^1Σ^+ \leftrightarrow b^3Π$ transition features a nearly diagonal Franck-Condon factor and has been proposed useful for probing and manipulating the ultracold molecular gas. We measure the transition strength directly by probing the ac Stark shift induced by near resonance light and determine the total excited-state spontaneous emission rate by observing the loss of molecules. From the extracted branching ratio and the theoretical modeling, we find that the leakage to the continuum of the $a^3Σ^+$ state plays the dominant role in the total transition linewidth. Based on these results, we show that it is feasible to create optical trapping potentials for maximizing the rotational coherence with laser light tuned to near this transition.

physics.atom-ph↗
arXiv2020

Observation of resonant dipolar collisions in ultracold $^{23}$Na$^{87}$Rb rotational mixtures

We report the investigation on dipolar collisions in rotational state mixtures of ultracold bosonic $^{23}$Na$^{87}$Rb molecules. The large resonant dipole-dipole interaction between molecules in rotational states of opposite parities brings about significant modifications to their collisions, even when an electric field is not present. In this work, this effect is revealed by measuring the dramatically enhanced two-body loss rate constants in the mixtures. In addition, the dipolar interaction strength can be tuned by preparing the NaRb mixture in different rotational levels with microwave spectroscopy. When the rotational level combination is not of the lowest energy, contributions from hyperfine changing collisions are also observed. Our measured loss rate constants are in good agreement with a quantum close-coupling calculation which we also present in full detail.

physics.atom-ph↗
arXiv2020

Anisotropic Polarizability of Ultracold Ground-state $^{23}$Na$^{87}$Rb Molecules

We report measurements of the ac polarizabilities of ultracold ground-state $^{23}\rm{Na}^{87}\rm{Rb}$ molecules. While the polarizability of the ground rotational state $J = 0$ is isotropic, that of the first excited rotational state $J = 1$ is anisotropic and depends strongly on the light polarization angle. We obtain both polarizabilities precisely by combining trap oscillation frequency measurement and high resolution rotational spectroscopy driven by microwave. With the optimized light polarization angle and intensity combination, the nonuniformity of the differential ac Stark shift between the two rotational states is minimized and the rotational coherence time is observed to be the longest.

physics.atom-ph↗
arXiv2020

Black-box Adversarial Sample Generation Based on Differential Evolution

Deep Neural Networks (DNNs) are being used in various daily tasks such as object detection, speech processing, and machine translation. However, it is known that DNNs suffer from robustness problems -- perturbed inputs called adversarial samples leading to misbehaviors of DNNs. In this paper, we propose a black-box technique called Black-box Momentum Iterative Fast Gradient Sign Method (BMI-FGSM) to test the robustness of DNN models. The technique does not require any knowledge of the structure or weights of the target DNN. Compared to existing white-box testing techniques that require accessing model internal information such as gradients, our technique approximates gradients through Differential Evolution and uses approximated gradients to construct adversarial samples. Experimental results show that our technique can achieve 100% success in generating adversarial samples to trigger misclassification, and over 95% success in generating samples to trigger misclassification to a specific target output label. It also demonstrates better perturbation distance and better transferability. Compared to the state-of-the-art black-box technique, our technique is more efficient. Furthermore, we conduct testing on the commercial Aliyun API and successfully trigger its misbehavior within a limited number of queries, demonstrating the feasibility of real-world black-box attack.

cs.LG↗
arXiv2020

On static and hydrodynamic biaxial nematic liquid crystals

In the first part of this paper, we will consider minimizing configurations of the Oseen-Frank energy functional $E(n, m)$ for a biaxial nematics $(n, m):Ω\to \mathbb S^2\times \mathbb S^2$ with $n\cdot m=0$ in dimension three, and establish that it is smooth off a closed set of $1$-dimension Hausdorff measure zero. In the second part, we will consider a simplified Ericksen-Leslie system for biaxial nematics $(n, m)$ in a two dimensional domain and establish the existence of a unique global weak solution $(u, n, m)$ that is smooth off at most finitely many singular times for any initial and boundary data of finite energy. They extend to biaxial nematics of earlier results corresponding to minimizing uniaxial nematics by Hardt-Kindelerherer-Lin \cite{HKL} and a simplified hydrodynamics of uniaxial liquid crystal by Lin-Lin-Wang \cite{LLW10} respectively.

math.AP↗