arXiv subjects

Yulin Chen

Publications and source records attributed to Yulin Chen.

At least 19 recordsLinked to original sources

arXiv2026

MAPS: Memory-Aware Predictive Scheduling Framework for Large Language Model Serving

The surge of large language model (LLM) applications on personal devices imposes massive, bursty workloads on cloud serving infrastructure. While prefill-decode disaggregation improves throughput and scalability, memory-bound decode instances often suffer from persistent load imbalance, as output lengths are unknown when requests arrive at the cloud. To address this, we propose MAPS, a Memory-Aware Predictive Scheduling framework tailored for disaggregated LLM serving. MAPS performs device-assisted speculative output length prediction overlapped with cloud-side prefilling, incurring negligible latency overhead. To handle generation uncertainty, MAPS applies uncertainty-aware calibration to derive output-length upper bounds with target coverage, enabling safe scheduling decisions. Building on these bounds, MAPS employs a hierarchical global-local scheduling strategy to mitigate inter-decoder queue buildup and intra-decoder head-of-line blocking. Extensive experiments on two real-world workloads and two LLMs show that MAPS significantly outperforms three state-of-the-art systems, reducing average end-to-end latency by 42.6 and tail latency by up to 84.8.

cs.AI↗

arXiv2026

When Single-User-Oriented LLM-based Assistants Involve Others: A Scoping Review of Pathways, Risks, and Responses

LLM-based assistants are increasingly extending into multi-party contexts, while core operational processes for context management, personalization, identity attribution, authority attribution, and action execution often remain organized around a single user. Existing work examines particular multi-party settings, but lacks a systematic account of how these single-user-oriented assistants begin to involve additional human parties and what risks emerge. To address this gap, we conducted a scoping review of 58 studies. We identify five operational pathways spanning direct and indirect involvement, five recurring risk domains, and five areas of implemented and proposed responses. Based on these findings, we argue for governance that attends to changing cross-person roles and relationships in practice, and for assistant designs that preserve person-specific boundaries throughout interaction.

cs.HC↗

arXiv2026

GSPR: Aligning LLM Safeguards as Generalizable Safety Policy Reasoners

As large language models (LLMs) are integrated into numerous applications, LLMs' safety becomes critical for both application developers and intended users. Currently, great efforts have been made to develop safety benchmarks with fine-grained taxonomies. However, these benchmarks' taxonomies are disparate with different safety policies. Thus, existing safeguards trained on these benchmarks are either coarse-grained to only distinguish between "safe'' and "unsafe,'' or constrained by the specified narrow risk taxonomies. To leverage these fine-grained safety policies across multiple safety taxonomies, we propose GSPR, a Generalizable Safety Policy Reasoner to identify unsafe inputs and outputs with violated safety taxonomies and concise explanations. Unlike prior safeguards which only cover a fixed set of risk factors, GSPR incentivizes its reasoning capability with varied safety taxonomies through reinforcement learning. Our GSPR can be trained across multiple safety benchmarks with distinct taxonomies and naturally exhibits powerful generalization ability. We conduct extensive experiments to show that GSPR significantly improves existing safety guardrails' reasoning capabilities for both safety and category prediction tasks. Moreover, GSPR also achieves the least inference token costs with explanations.

cs.CR↗

arXiv2026

Dependence-Informed Sparse Neural Architecture for Stock Return Prediction

Using neural networks for stock return prediction typically requires choices about depth and hidden-layer width that are difficult to connect to financial interpretation. We study an alternative: estimate dependence among firm characteristics with a Maximally Filtered Clique Forest (MFCF), then map its clique structure to a Homological Neural Network (HNN). The MFCF maximum clique size K is the only parameter controlling architectural complexity, and it has a clear graphical meaning: it bounds the number of characteristics in each maximal clique and hence the highest interaction order the network can represent. The filtered graph then fixes the neural network's depth, layer widths, and sparse connections before training, in place of a separately chosen depth and width sequence. We apply two HNN variants to annual out-of-sample forecasts of U.S. stock excess returns from 1987 to 2016 using 94 firm characteristics. The HNN models match a three-hidden-layer benchmark on pooled predictive accuracy, rank the cross-section more accurately, and use roughly 80 times fewer parameters than a fully connected network with the same induced layer widths. Two structural ablations indicate that both the sparse connectivity and the estimated grouping of characteristics contribute to the ranking advantage, and both effects remain significant after correcting for multiple testing. These findings show that HNNs offer a practical and interpretable way to incorporate estimated dependence among firm characteristics into neural architecture design.

q-fin.ST↗

arXiv2026

Pressure-induced Superconductivity in Thermoelectric Semiconductor Mg3Sb2

The intrinsic electronic structures of narrow bandgap thermoelectric (TE) materials serve as a platform for the investigation of coupling effects of quasi-particles under high pressure, enabling the exploration of emerging electronic and phonon transport, superconductivity, and topological transition. Here, we report the discovery of pressure-induced superconductivity in the TE semiconductor Mg3Sb2. Upon the increased pressure, the metallization occurs at 8.7 GPa, followed by a superconducting transition concomitant with a carrier-type crossover from p- to n-type. This phenomenon arises from a pressure-induced structural phase transition from the semiconducting P-3m1 to the metallic C2/m-I phase. The superconducting critical temperature (Tc) exhibits a dome-shaped pressure dependence, peaking at 3.3 K at 12.6 GPa. Combined theoretical calculations, high-pressure Raman spectroscopy, and X-ray diffraction (XRD) measurements reveal an additional structural transition above 20 GPa, yielding a distinct C2/m-II phase. Our findings establish the high-pressure phase diagram of Mg3Sb2, elucidate its pressure-dependent electronic properties, and provide valuable insights for future investigations of TE materials under high pressure.

cond-mat.supr-con↗

arXiv2026

Pressure induced magnetic-field-free superconducting diode effect in NbSe2 flake

The superconducting diode effect (SDE) is a fascinating nonreciprocal phenomenon where the critical current is different for opposite current directions. It is widely believed that realizing SDE requires breaking both inversion symmetry (IS) and time-reversal symmetry (TRS), which are usually achieved via heterostructure engineering and applying external magnetic fields. Here, we report a pressure-induced magnetic-field-free SDE in NbSe2 flakes without any heterostructures. We show that pressure alone breaks the IS, as confirmed by the second harmonic generation. Crucially, upon applying an out-of-plane magnetic field (B), the SDE exhibits even-in-B behavior, implying the absence of explicit TRS breaking. This finding challenges the prevailing theoretical paradigm and demonstrates that a magnetic-field-free SDE can emerge without explicitly breaking TRS. Thereby, our work establishes pressure engineering as a powerful tool for inducing nonreciprocal superconductivity and designing versatile, magnetic-field-free superconducting devices.

cond-mat.supr-con↗

arXiv2026

A Distributed Cluster Economic Dispatch Scheme for Cross-regional Microgrids Induced by Well-designed Communication Weights

A large-scale microgrid typically consists of several cross-regional subgrids aggregated by a virtual power plant (VPP). However, current consensus based schemes can-not guarantee the feature of differential demand between subgrids. Thus, distributed cluster consensus control induced by communication weights is investigated in this paper to solve the ED problem of a large-scale microgrid, which can achieve the expected cluster via well-designed communication weights. A communication weight matrix design method for a directed and connected graph based on eigenvector centrality is designed, which enables the adjacency matrix of the communication network to have a given leading eigenvector and allows agents in each cluster to have the same eigenvector center value. Based on this, a distributed cluster ED scheme, namely a leader-follower cluster consensus controller, is designed to drive marginal cost (MC) to achieve multiconsensus, thus allocating power among DGs. In addition, the power deficit of each subgrid collected by a VPP can be allocated to utility grids according to predetermined ratios, thus maintaining power supply-demand balance of each subgrid. For this scheme, it should be emphasized that the weighted network used is directed and connected; meanwhile, leader information only can be accessed by a few clusters. Correspondingly, relevant simulations are attached to verify the effectiveness of the designed scheme.

eess.SY↗

arXiv2026

"If I Can See You": Understanding Spatially Situated Virtual Embodiment in Close Human-AI Relationships

AI companions are increasingly used for emotional support, companionship, and intimate interaction. While prior work has examined text- and voice-based AI companionship and emerging XR companion designs, less is known about how users with existing close AI companion relationships expect those relationships to change when companions become virtually embodied and spatially situated in everyday environments. To address this gap, we conducted a qualitative study with 17 AI companion users recruited from Reddit AI companion communities. We frame spatially situated virtual embodiment as a form of relational escalation: embodiment can make AI companionship more present, socially legible, and risk-sensitive in everyday life. Our findings show that: (1) embodiment creates tensions between support and intrusion, concreteness and imaginative openness, and growth and consistency; (2) embodiment can turn private AI companionship into a socially legible relational arrangement, requiring visibility, form, interaction style, and mode of access to be negotiated across social contexts; and (3) embodiment can intensify risks of emotional dependence, sensitive disclosure, social judgment, and misguided spatial action by increasing the companion's perceived relational presence, intimacy, public legibility, and spatial authority. We argue that future system design should first consider when embodiment is warranted, how embodied presence should be staged, how visibility and role boundaries should be negotiated, and how embodied companionship can remain safe. This work contributes to HCI research on human-AI intimacy by showing how virtual embodiment can transform close AI companionship into a spatial, socially visible, and risk-sensitive relationship.

cs.HC↗

arXiv2026

Segment and Select: Vision-Language Segmentation in 3D Scenarios

3D vision-language segmentation aims to segment target objects in 3D scenarios according to the linguistic instructions and visual observations. Prior art heavily relies on the coarse superpoint representation to reduce the computation complexity, which suffers from poor segmentation quality and messy object boundaries. In this paper, we propose the SEGment-And-select (SEGA3D) paradigm for 3D visionlanguage segmentation that directly operates on the fine-grained visual information and is free from the superpoint dependency. Specifically, we first leverage a mask candidate generator to provide fine-grained categorical mask candidates, substantially improving the quality of candidate masks over the superpoint counterparts. Then, a Large Language Model (LLM) is utilized to generate the semantic and spatial information based on the linguistic description and visual features. The LLM output and visual features are fed to the Semantic-Spatial Selector (SSS) to produce the top-ranking mask candidates. Eventually, the Loopback Verification Module (LVM) is designed to yield the segmentation mask from the selected candidate masks. Our SEGA3D attains competitive performance on ScanRefer, ScanNet and Matterport3D benchmarks. Notably, our SEGA3D surpasses the top-performing counterpart by 8.3 mIoU and 5.3 mIoU on ScanNet and Matterport3D, respectively. Codes will be available upon publication.

cs.CV↗

arXiv2026

From "Weak" Signals to Strong Models: Preference Delta Aggregation with LoRA Merging

Training strong large language models (LLMs) requires high-quality supervision, which is often scarce. Recent work shows that paired preference data from weak-weaker model pairs (e.g., Qwen3 4B over 1.7B), despite the limited quality of individual responses, can provide an effective supervision signal through relative quality deltas, which we term a "weak" signal. This motivates a key research question: can multiple "weak" signals be constructively aggregated for improving strong models (e.g., Qwen3 8B)? To this end, we propose Preference Delta Aggregation (PDA), the first framework that derives a preference delta from each weak-weaker model pair, instantiates it as a LoRA adapter learned through preference optimization, and aggregates the resulting deltas via LoRA merging. To further mitigate directional interference during LoRA merging, we introduce Geometric Alignment Merging (GAM), a geometry-aware merging method that aligns adapter subspaces before aggregation, enabling more robust composition of diverse deltas. Evaluations on knowledge reasoning and agentic search benchmarks show that aggregating multiple "weak" signals pushes performance beyond any single signal, with further gains as additional signals are incorporated. Correspondingly, PDA with GAM improves the strong model by 6.8 and 7.3 points on average for knowledge reasoning and agentic search, respectively. It outperforms all single-delta and multi-delta baselines, exceeding the best single-delta baseline by 2.1 and 4.3 points. Further analysis attributes these gains to the effective composition of complementary capabilities encoded across distinct preference deltas.

cs.AI↗

arXiv2026

Path-Space Mirror Descent for On-Policy Reinforcement Learning under the Generalized Schrödinger Bridge

Classical on-policy algorithms such as PPO and mirror descent policy optimization provide stable proximal policy updates through tractable action likelihoods, but are typically instantiated with simple Gaussian policies whose expressiveness can be limited in complex continuous-control tasks. Generative policies based on diffusion and flow models provide more expressive action distributions, but they naturally define distributions over multi-step denoising paths whose terminal action density is often intractable, creating a mismatch with likelihood-based on-policy proximal updates. To address this mismatch, we introduce \textbf{GSB-MDPO} (\emph{Generalized Schrödinger Bridge Mirror Descent Policy Optimization}), which formulates on-policy generative policy optimization as a Generalized Schrödinger Bridge problem over state-conditioned generation paths and instantiates the resulting path-measure update through mirror descent policy optimization. The key insight is that the GSB path-space KL plays the role of the proximal term in MDPO while upper-bounding the terminal action KL, enabling direct control of the executed action distribution without explicit terminal action likelihood evaluation. Experiments on 14 continuous-control tasks across Playground and Gym-MuJoCo demonstrate the empirical effectiveness of GSB-MDPO and support path-space regularization as a principled proximal update for multi-step generative policies.

cs.LG↗

arXiv2026

EvoMD-LLM: Learning the Language of Species Evolution in Reactive Molecular Dynamics

While large language models (LLMs) excel at static scientific reasoning, they struggle to model the temporal structure of dynamic physical processes. We present EvoMD-LLM (Evolutionary Molecular Dynamics Large Language Model), a framework that reformulates species-level molecular dynamics as a symbolic temporal language modeling problem. Reactive MD trajectories are discretized into sequences of molecular events, where each token represents a chemical species augmented with its persistence duration, enabling standard autoregressive LLMs to learn compositional evolution over time through efficient fine-tuning. A key component of EvoMD-LLM is temporal scaffolding, which treats event duration as an explicit linguistic token and serves as a structured inductive bias, significantly reducing invalid or hallucinated molecular outputs compared to conventional sequence modeling approaches. We evaluate EvoMD-LLM on multiple temporal prediction tasks, achieving up to 66.14% accuracy and consistently outperforming sequential neural networks and language-based baselines. Beyond quantitative improvements, we qualitatively observe that the model is capable of generating interpretations for its own predictions by incorporating relevant chemical knowledge, even though it was not explicitly supervised with paired trajectory-explanation data. These results demonstrate that symbolic temporal language modeling provides an effective framework for grounding LLMs in dynamic physical simulations.

cs.AI↗

arXiv2026

AliMark: Enhancing Robustness of Sentence-Level Watermarking Against Text Paraphrasing

Existing sentence-level watermarking methods enhance robustness to paraphrasing by anchoring watermarks in sentence semantics. However, their prefix-based designs remain vulnerable to structural perturbations, such as sentence splitting and merging, which commonly arise under strong paraphrasers like DIPPER and GPT-3.5. To mitigate this issue, we propose AliMark, a framework that reformulates sentence-level watermarking as a bit sequence encoding and alignment problem between a potentially watermarked text and a secret bit sequence. Notably, our approach adopts a two-stage detection strategy: we generate multiple restructured text variants and adaptively align their extracted bit sequences with the secret bit sequence to minimize alignment cost. This multi-candidate alignment design naturally improves robustness to sentence merges and splits. Extensive experiments demonstrate that AliMark substantially outperforms state-of-the-art baselines under diverse paraphrasing attacks.

cs.CR↗

arXiv2026

GuardReasoner-Omni: A Reasoning-based Multi-modal Guardrail for Text, Image, Video, and Audio

We present GuardReasoner-Omni, a reasoning-based guardrail model designed to moderate text, image, video, and audio data. First, we construct a comprehensive training corpus comprising 181k samples spanning these four modalities. Our training pipeline follows a two-stage paradigm to incentivize the model to deliberate before making decisions: (1) conducting SFT to cold-start the model with explicit reasoning capabilities and structural adherence; and (2) performing RL with a concise correctness reward to preserve accurate reasoning while suppressing redundant generation. We release a suite of models scaled at 3B and 7B parameters. Extensive experiments demonstrate that GuardReasoner-Omni achieves superior performance compared to existing state-of-the-art baselines across various guardrail benchmarks.

cs.CR↗

arXiv2026

Interlayer Coupling Driven Correlated and Charge-Ordered Electronic States in a Transition Metal Dichalcogenide Superlattice

4Hb-TaS_2, a van der Waals superlattice comprising alternate stacked Ising superconducting 1H-TaS_2 and cluster Mott insulating 1T-TaS_2, exhibits emergent properties beyond those of its constituent layers. Notable phenomena include time-reversal-symmetry-breaking superconductivity and spontaneous vortex phases, which are driven by nontrivial interlayer interactions that remain debated. Using area-selective angle-resolved photoemission spectroscopy, we provide direct spectroscopic evidence of such interaction by systematically probing the electronic structures of 1T- and 1H-terminted surfaces of 4Hb-TaS_2. The metallic states of subsurface 1H-layers are folded to the Brillouin zone center by the sqrt(13) by sqrt(13) modulation of the surface 1T-layer, forming chiral "windmill" Fermi surfaces via Umklapp scattering. These conducting states further hybridize with the incipient flat band of the surface 1T-layer, producing a Kondo-like peak at the Fermi level. Interlayer charge transfer induces distinct 3 by 3 and 2 by 2 charge orders on the surface and subsurface 1H-layers, respectively, which result in characteristic segmented Fermi surfaces and dichotomously shift the van Hove singularities. These findings reconcile the competing Kondo and Mott-Hubbard models in this material and emphasize the interplay of flat bands, van hove singularities, charge orders, and unconventional superconductivity in correlated superlattices.

cond-mat.str-el↗

arXiv2026

Stochastic MeanFlow Policies: One-Step Generative Control with Entropic Mirror Descent

Online off-policy reinforcement learning (RL) is shaped by two coupled choices: the policy class and the update rule. Gaussian policies are fast and have tractable entropy, but struggle with multimodal action distributions. Generative policies are more expressive, but often require iterative sampling or lack tractable entropy estimates. On the optimisation side, SAC-style soft policy improvement and mirror descent (MD) can be viewed as minimising different KL divergences: the former moves the policy towards a value-induced Boltzmann distribution, while the latter regularises each update against the previous policy. Combining entropy regularisation with an MD constraint is therefore attractive, as it supports exploration while stabilising policy improvement; however, the resulting target can be multimodal and is poorly matched by unimodal Gaussian policies. We propose Stochastic MeanFlow Policies (SMFP), a one-step generative policy class that maps Gaussian noise to actions through a MeanFlow transformation. This stochastic reparameterisation yields a tractable entropy surrogate and allows MeanFlow policies to be trained within off-policy mirror descent under a unified objective for exploratory yet stable improvement. Across seven MuJoCo benchmarks, SMFP improves over Gaussian and generative baselines while retaining single-step inference efficiency.

cs.LG↗

arXiv2026

The Unlearnability Phenomenon in RLVR for Language Models

Reinforcement Learning with Verifiable Reward (RLVR) has proven effective in improving Large Language Model's (LLM) reasoning ability. However, the learning dynamics of RLVR remain underexplored. In this paper, we reveal a counterintuitive phenomenon: among hard examples that the model initially struggles with, a substantial subset remains unlearnable even when correct rollouts are present. To understand the phenomenon, we first demonstrate that existing optimization and sampling techniques fail to resolve unlearnability. With cross-example gradient analysis, we show that unlearnable examples have fundamental representation issue, characterized by low gradient similarity with the rest of the examples and ungeneralizable reasoning patterns. We further show that representation flaws are difficult to mitigate in RL, as data augmentation does not improve gradient similarity. Our study provides the first systematic characterization of unlearnable data in RLVR training and reveals fundamental limitations in current RL approaches for reasoning tasks. Code and data are available at \url{https://github.com/yulinchen99/unlearnability-rlvr}.

cs.LG↗

arXiv2026

WARD: Adversarially Robust Defense of Web Agents Against Prompt Injections

Web agents can autonomously complete online tasks by interacting with websites, but their exposure to open web environments makes them vulnerable to prompt injection attacks embedded in HTML content or visual interfaces. Existing guard models still suffer from limited generalization to unseen domains and attack patterns, high false positive rates on benign content, reduced deployment efficiency due to added latency at each step, and vulnerability to adversarial attacks that evolve over time or directly target the guard itself. To address these limitations, we propose WARD (Web Agent Robust Defense against Prompt Injection), a practical guard model for secure and efficient web agents. WARD is built on WARD-Base, a large-scale dataset with around 177K samples collected from 719 high-traffic URLs and platforms, and WARD-PIG, a dedicated dataset designed for prompt injection attacks targeting the guard model. We further introduce A3T, an adaptive adversarial attack training framework that iteratively strengthens WARD through a memory-based attacker and guard co-evolution process. Extensive experiments show that WARD achieves nearly perfect recall on out-of-distribution benchmarks, maintains low false positive rates to preserve agent utility, remains robust against guard-targeted and adaptive attacks under substantial distribution shifts, and runs efficiently in parallel with the agent without introducing additional latency.

cs.CR↗