Search arXiv⌕ Search

arXiv subjects

Shuang Liang

Publications and source records attributed to Shuang Liang.

At least 37 records · Page 2Linked to original sources
arXiv2026

LEEPS: Latent-Guided Explore-Exploit Prompt Sampling for Efficient RLVR in Large Language Models

Reinforcement learning with verifiable rewards (RLVR) improves the reasoning capabilities of large language models, but prompt groups with identical rollout rewards consume generation budget without effective learning signals. Pre-rollout prompt selection can reduce this waste by screening prompts before rollout generation. However, existing pre-rollout methods struggle to balance exploitation and exploration: repeatedly exploiting historically informative prompts can narrow training coverage, whereas broader exploration can lower the fraction of informative prompts. To address these limitations, we introduce LEEPS, a Latent-Guided Explore--Exploit Prompt Sampler that adaptively balances the reuse of previously observed informative prompts with continued exploration of uncertain ones. LEEPS partitions candidates into exploit and explore portfolios and adaptively allocates rollout budget according to their recent non-trivial ratios. It further uses representation-space neighbors and historical rollout outcomes to prioritize uncertain prompts likely to yield non-zero reward variance, thereby making exploration more targeted without additional rollouts. Across six mathematical reasoning benchmarks, LEEPS achieves the highest average score at both model scales, with relative gains of 2.6\% and 3.7\% over the strongest baseline for Qwen2.5-Math-1.5B and 7B, respectively, and generally improves faster during the training process. It also achieves the highest average score across the three evaluated OOD general-reasoning benchmarks at both model scales and adds only about 2 seconds of online sampling overhead per training step. Code is available at https://github.com/ShuangLiangX/LEEPS.

cs.CL↗
arXiv2026

Visual Graph Reasoning via Knowledge Compilation

Visual graph reasoning requires answering graph-theoretic questions directly from graph images, where graph topology and state are conveyed visually rather than given in symbolic form. Despite recent progress of vision-language models (VLMs), current approaches to visual graph reasoning still fail on simple visual graph problems. This reveals a fundamental limitation of existing approaches: they prioritize final-answer supervision over the intermediate recovery of an explicit graph representation that preserves graph topology and state from visual input. To address this limitation, we propose VGCompiler, a compilation-centric paradigm for visual graph reasoning via knowledge compilation. VGCompiler organizes reasoning around two compilers: a representation compiler that recovers a structure-preserving intermediate graph representation from visual input, and an operation compiler that compiles query intent under the recovered graph state into an executable graph operation. Specifically, we build VGCompiler on Qwen3-VL-8B and train it with reinforcement learning guided by a layered reward over executability, compiled graph validity, representation quality, and operation quality. VGCompiler uses a frozen observer to summarize graph and question conditions into lightweight signatures, enabling archive retrieval and code reuse across similar regimes. Experiments on three benchmarks GVLQA, VisionGraph, and VGCURE, show that Qwen-VGCompiler, built on an 8B backbone, surpasses the strongest closed-source VLM baseline by 28.9% and the strongest code-based baseline by 23.7%, while maintaining high efficiency. We further evaluate VGCompiler on three real-world domains, including metro routing, logistics delivery, and network fault assessment, where it generalizes across heterogeneous visual graphs and domain-grounded tasks.

cs.MM↗
arXiv2026

DeepStack: Facilitating Co-Design Exploration of 3D DRAM-Stacked Accelerators for Distributed LLM Inference

Advances in hybrid bonding and packaging have driven growing interest in 3D DRAM-stacked AI accelerators. As large language models (LLMs) scale to hundreds of billions or trillions of parameters, distributed inference across multiple 3D chips has become essential for AI serving. This trend makes cross-stack co-design critical because system-level parallelization and scheduling choices are tightly coupled with hardware characteristics such as memory organization, interconnects, and thermal constraints. We present DeepStack, an accurate performance model and efficient design space exploration (DSE) framework for distributed 3D-stacked LLM inference. At the hardware level, DeepStack captures transaction-aware memory bandwidth, bank activation constraints, buffering limitations, and thermal and power behavior. At the system level, it incorporates comprehensive parallelization strategies and execution scheduling. Through a dual-stage network abstraction and tile-level compute-communication overlap modeling, DeepStack achieves up to 100,000x faster evaluation than state-of-the-art simulators at comparable accuracy. We cross-validate DeepStack against our in-house 3D designs, an NS-3 backend with 2.12% error, and vLLM serving on eight B200 GPUs with 12.92% error. Combined with hierarchical search, DeepStack efficiently explores about 2.5 x 10^14 design points spanning the number of stacked DRAM layers, DRAM vertical connectivity, interconnects, compute-memory allocation, and distributed scheduling under thermal and area constraints. A search-space ablation shows that restricted DSE baselines can miss up to 9.5x modeled throughput. Beyond modeling and DSE, DeepStack derives design implications for distributed 3D AI systems and guides performance optimization across the stack. Source code and artifacts are available at https://github.com/tile-ai/DeepStack/tree/ae.

cs.AR↗
arXiv2026

Probabilistic characterization of blending with LSST and application to cluster lensing cosmology

Next-generation galaxy surveys, like the Vera C. Rubin Observatory's Legacy Survey of Space and Time (LSST), will deliver unprecedented depth and sky coverage, enabling precise measurements of cosmic probes such as weak lensing and galaxy clustering. However, increased imaging depth leads to significant blending of galaxy images, particularly in dense fields like galaxy clusters. This blending, exacerbated by atmospheric blurring in ground-based observations, contaminates galaxy property measurements and causes source confusion. To simultaneously capture these effects, we develop a probabilistic framework introducing the blending entropy, a metric quantifying the ambiguity in matching detected objects to true galaxies or external reference sources. Using simulated data from the DESC Data Challenge 2 (DC2), we characterize blending in LSST data and quantify its impact on cluster lensing cosmology around cosmoDC2 halos. We demonstrate that imposing a blending entropy threshold of $S_b<0.2$ effectively filters out highly blended objects (around 25%), which are especially prevalent near the survey's magnitude limit and are associated with higher errors in shape measurements and photometric redshifts. Applying this cut substantially reduces blending-induced biases in cluster lensing profiles and mass estimates, thereby mitigating systematic errors in cosmological parameters---most notably reducing tension in $σ_8$ estimates. Our method is readily generalizable to other static probes and offers a practical path forward for real data analyses, particularly when leveraging overlapping high-resolution datasets from spaced-based missions such as Euclid or the Roman Space Telescope, where these external datasets can act as reference catalogs to improve the identification of blended sources in LSST data.

astro-ph.CO↗
arXiv2026

GUT: Quantifying and Optimizing the Reasoning Uncertainty of LLMs via Graph Complexity

Recent years have witnessed great advances in the reasoning ability of Large Language Models (LLMs). However, the reasoning processes of LLMs often exhibit uncertainty, where LLMs often produce a proliferation of divergent branches at each reasoning step even when fed the same prompting inputs, and certain branches exhibit evidently incredible, even nonsensical, reasoning chains and results. In this paper, we propose the Graph-complexity-based UncerTainty (GUT) method for investigating the reasoning uncertainty of LLMs. The key idea of GUT is to characterize the potential branches of each reasoning chain with a directed acyclic graph, thereby ensuring that all potential branches are comprehensively covered within the graph space. Building upon this recognition, we further build two modules of GUT, that is, a Quantification (GUT-Q) module and an Optimization (GUT-O) module, for quantifying and reducing the reasoning uncertainty of LLMs, respectively. GUT-Q measures LLM reasoning uncertainty by approximating the reasoning space complexity with graph complexity. GUT-O implements uncertainty optimization by treating negative uncertainty as the reward function in reinforcement learning. Experimental results conducted on four LLMs and five datasets validate the effectiveness of GUT.

cs.AI↗
arXiv2026

Learning to Detect Unseen Jailbreak Attacks in Large Vision-Language Models

Despite extensive alignment efforts, Large Vision-Language Models (LVLMs) remain vulnerable to jailbreak attacks. To mitigate these risks, existing detection methods are essential, yet they face two major challenges: generalization and accuracy. While learning-based methods trained on specific attacks fail to generalize to unseen attacks, learning-free methods based on hand-crafted heuristics suffer from limited accuracy and reduced efficiency. To address these limitations, we propose Learning to Detect (LoD), a learnable framework that eliminates the need for any attack data or hand-crafted heuristics. LoD operates by first extracting layer-wise safety representations directly from the model's internal activations using Multi-modal Safety Concept Activation Vectors classifiers, and then converting the high-dimensional representations into a one-dimensional anomaly score for detection via a Safety Pattern Auto-Encoder. Extensive experiments demonstrate that LoD consistently achieves state-of-the-art detection performance (AUROC) across diverse unseen jailbreak attacks on multiple LVLMs, while also significantly improving efficiency. Code is available at https://github.com/ShuangLiangX/Learning-to-Detect.

cs.CR↗
arXiv2026

Nonlinear Diamagnetic Interactions in Ultrastrongly Coupled 2D Electrons

The quantum Hopfield model is widely used to describe ultrastrong light--matter coupling between cavity photons and collective bosonic excitations in solids, where the diamagnetic interaction is conventionally assumed to be a constant. We experimentally demonstrate that the diamagnetic response of Landau polaritons is reduced under strong terahertz field excitation. We show that this behavior originates from field-driven redistribution of electrons into the nonparabolic regime of the conduction band of GaAs, which reduces the plasma frequency and consequently the diamagnetic interaction strength. A microscopic hot-electron model reproduces the observed nonlinear response. Motivated by this microscopic picture, we propose a nonlinear extension of the Hopfield model with a Kerr-like interaction. Our results establish a route toward nonlinear cavity quantum electrodynamics and driven ultrastrong light--matter coupling beyond the conventional linear Hopfield description, which is capable of creating uniquely quantum optical effects such as squeezed light generation.

quant-ph↗
arXiv2026

AcceptMoE: Commitment-Weighted Self-Sizing Verifier Expert Sets for Efficient MoE Speculative Decoding

Speculative decoding verifies a tree of draft tokens in one target-model forward pass. For a mixture-of-experts (MoE) target, however, parallel verification can activate the union of the experts selected by all tree nodes, even though only a small subset of those nodes reaches the accepted output. Token count, activated-expert union size, and expert-weight traffic are therefore distinct cost measures: reducing the token workload need not shrink the expert union proportionally, and under offloading, transfer traffic also depends on cache residency. We introduce AcceptMoE, a verifier-side expert selector that combines target-router scores with offline-estimated commitment probabilities and automatically adjusts the number of eligible experts for each verification block, eliminating the need for a user-specified expert budget. Under offloading, AcceptMoE conditions expert eligibility on cache residency instead of predicting natural routes and prefetching the corresponding expert weights. Although constraining target-expert eligibility changes the model distribution, across 12 model-task pairs spanning three MoE targets and four benchmarks, AcceptMoE's mean accuracy is 0.27 percentage points lower than that of EAGLE-3 speculative decoding with natural routing. Served with SGLang at batch size one, it reaches 1.290 times the throughput of this baseline with all expert weights in GPU memory, and 2.06 times under physical expert offloading, while reducing host-to-device traffic by 73.6 percent to 77.1 percent.

cs.LG↗
arXiv2026

When Safety Becomes a Vulnerability: Exploiting LLM Alignment Homogeneity for Transferable Blocking in RAG

Retrieval-Augmented Generation (RAG) systems are vulnerable to blocking attacks, in which poisoned documents cause large language models (LLMs) to refuse benign queries. Existing attacks rely on adversarial suffixes or explicit instructions, which are increasingly ineffective against modern LLMs, susceptible to prompt injection filtering, or require feedback from the target system. We observe overlapping risk categories and refusal criteria across safety-aligned LLMs, a phenomenon we term alignment homogeneity. This shared attack surface makes refusal-inducing context transferable across models. Accordingly, we propose TabooRAG, which optimizes one document per query for retrieval and refusal induction in a surrogate RAG environment, then transfers it to an unknown target system. Rather than injecting instructions, TabooRAG constructs query-relevant risk context to trigger alignment-driven refusal. To reduce optimization cost, it reuses validated strategies through a query-aware strategy library. Across nine LLMs and three datasets, TabooRAG achieves state-of-the-art ASR after filtering, with a 67.3% relative gain over the average per-setting best baseline. Further experiments show that TabooRAG remains effective with diverse surrogate models, against unseen target models, and under stronger RAG pipelines and existing defenses.

cs.CR↗
arXiv2026

OpsLLM: Construction of Large Language Model for Software Operations with Multi-stage Learning

In the field of software operations, Large Language Models (LLMs) have attracted increasing attention. However, existing research has not yet achieved efficient and effective endto-end intelligent operations due to low-quality data, fragmented knowledge and insufficient learning. To explore the potential of LLMs in software operations, we propose OpsLLM, a domainspecific LLM that supports both knowledge-based question answering (QA) and root cause analysis (RCA). Moreover, we disclose the detailed workflow for building LLMs specifically in the software operations domain. First, a Human-in-the-Loop mechanism is introduced to curate high-quality data from a large collection of operational data and construct a fine-tuning dataset. Then, based on the data, supervised fine-tuning is conducted to achieve a base model. Furthermore, we introduce a domain process reward model (DPRM) during the reinforcement learning stage to optimize the accuracy and reliability of the fine-tuned model on RCA tasks. Experimental results on the tasks with diverse difficulties demonstrate that OpsLLMs effectively learns and aligns with the operational domain knowledge infused, outperforming existing open-source and closed-source LLMs in accuracy with improvements of 0.2%~11.9% on QA tasks and 8.5%~70.3% on RCA tasks, while exhibiting strong transferability. Moreover, we will open-source three versions of OpsLLM with 7B, 14B and 32B parameters, along with a 15K fine-tuning dataset.

cs.LG↗
arXiv2026

Topological magnon noises

We develop a comprehensive formalism for magnon transport in ferromagnetic insulators driven by a temperature gradient. The formulas for magnon current and corresponding magnon noise are derived herein based on the spin Hamiltonian of a topological magnon insulator, which enables us to calculate the magnon Hall angle, to provide an explicit expression for the Fano factor, and to reaffirm the quantitative relations between magnon conductance and magnon noise. We find that the magnon current is not conserved in the presence of the Gilbert damping. Consequently, the reciprocal relation between the local and nonlocal noises, the Johnson-Nyquist formula between the conductance and the thermal noise, and the relation between the transmission coefficient and the shot noise are profoundly altered.

cond-mat.mes-hall↗
arXiv2026

Implicit Bias of SGD in Multivariate ReLU Networks: Effective Width Collapse

We study the implicit bias of noisy stochastic gradient descent in training wide two-layer ReLU networks for multivariate regression. In a mean-field regime, the training dynamics are approximated by a Wasserstein gradient flow that converges to a unique stationary measure. We characterize the structure of this stationary measure and the predictor it represents. We show that, despite the network being infinitely overparameterized, the learned predictor admits an effectively finite representation: the input weights and biases align along finitely many directions, leading to an effective width collapse. In particular, the solution function is continuous piecewise affine, with affine regions determined by the cells of a finite hyperplane arrangement. The number of learned directions, and hence hyperplanes, is bounded above by $2\mathcal{P}-1$, where $\mathcal{P}$ denotes the number of linear dichotomies realizable on the training inputs. We further establish a non-redundancy property of the learned representation by proving that each learned direction induces a unique ternary activation pattern on the training data. Consequently, the complexity of the learned predictor is governed by the combinatorial geometry of the training data.

cs.LG↗
arXiv2026

Mirror-mediated long-range coupling and robust phase locking of spatially separated exciton-polariton condensates

Lattice arrays are valuable simulators for complex mathematical problems, but physical systems typically allow only short-range coupling. We demonstrate a method for independently tunable, long-range interactions between polariton condensates in two-dimensional lattices by using vertical emission and external imaging to couple arbitrary sites. Two geometrically isolated condensates are phase-locked without planar coupling, verified via phase-resolved interferometry. Analytical modeling reveals mechanisms for robust coherence. The mirror-based scheme, free of cameras or modulators, offers a pure, high-bandwidth analogt element. Extension to dense graphs via segmented micro-mirrors is limited only by imaging optics, enabling scalable, energy-efficient polaritonic hardware for neuromorphic computation.

physics.optics↗
arXiv2026

LWDrive: Layer-Wise World-Model-Guided Vision-Language Model Planning for Autonomous Driving

Vision-Language Models (VLMs) provide powerful semantic understanding and commonsense reasoning for End-to-End Autonomous Driving (E2E-AD) planning. However, trajectories directly generated by VLMs often encode only coarse driving intentions and remain insufficient for geometrically accurate, future-aware, and multi-view-grounded planning. To address these limitations, we develop the Layer-Wise World-Model-Guided Driving framework (LWDrive). LWDrive is a VLM planning framework that refines coarse trajectories through layer-wise world-model guidance. Instead of treating the VLM output as the final trajectory, LWDrive uses it as an intent-aware coarse plan, expands a diverse candidate space around it, and progressively refines the candidates through a Foresight Cascade Planner (FCP). Specifically, we introduce future-frame generation supervision to encourage the VLM to learn forward-looking scene representations, thereby injecting planning-relevant predictive dynamics into its internal hidden states. Built upon these world-model-supervised representations, FCP exploits VLM features across multiple layers and integrates historical temporal states, Action-Query representations, and current-frame multi-view Bird's-Eye-View (BEV) features to refine candidate trajectories in a coarse-to-fine manner. This design enables progressive correction of spatial positions and motion trends while grounding trajectory refinement with multi-view scene cues and preserving the high-level driving intention produced by the large model. Finally, a score head evaluates the refined candidates and selects the best trajectory as the final planning output. Experiments show that LWDrive achieves a score of 92.0 on the NAVSIM benchmark and 89.6 on NAVSIM-v2. Code and models will be made publicly available.

cs.CV↗
arXiv2026

The On-Sky Performance of the LSST Camera CCD Array

The focal plane of the LSST Camera contains 189 individual science CCDs, arranged into 21 raft tower modules, along with 4 wavefront and 8 guider CCDs located in 4 additional corner RTMs. Altogether, the LSST Camera CCDs compose the largest focal plane ever constructed. The LSST Camera is the primary instrument of Rubin Observatory, which will begin the Legacy Survey of Space and Time in 2026. In this paper, we describe the on-sky performance of the LSST Camera CCDs, from receipt at NSF/DOE Vera C. Rubin Observatory in May 2024 to on-sky observations during the first year of operations. We discuss the process to establish functionality of several CCDs which were affected by an electrical short and faulty analog-digital converter, optimizations of readout timing in response to changes in the survey strategy, and implementation of enhanced focal plane safety measures through an active clearing mechanism on the CCDs. Finally, we discuss sensor features observed on-sky, and global performance during the first year of operations. The operations to date of the LSST Camera CCDs have demonstrated the capability of performing a wide, fast, and deep optical imaging survey of the entire southern sky at the Rubin Observatory.

astro-ph.IM↗
arXiv2026

An overview of stray light findings and interpretation during on-sky commissioning of LSSTCam

Wide-field telescopes are intrinsically difficult to shield from unwanted stray and scattered light, while the search to identify sources of contaminating light is frequently a challenging task. The Vera C.~Rubin Observatory, which achieved its first photon with the LSST Camera (LSSTCam) on April 15, 2025, will initiate a revolutionary era for the study of dark matter, dark energy, the transient sky, the Solar System, and the Milky Way. LSSTCam will provide near seeing-limited images of the sky in six bands ($u,g,r,i,z,y$) over a $3.^\circ 5$-diameter field of view, and over the course of a decade, it will execute the Legacy Survey of Space and Time (LSST). This work provides an overview of the dedicated stray and scattered light test campaign that has been undertaken since the start of Rubin commissioning. In particular, we highlight the processes used to characterize, model, and mitigate stray light present in LSSTCam images. The Rubin commissioning team created a series of testing and analysis tools to track stray light artifacts from their initial discovery through reproduction with timely observations, simulation using ray tracing to identify opto-mechanical origins, and finally devising corrective actions. The complex stray light features encountered by Rubin provide a wealth of experience for the future wide-field and extremely wide-field observatories. This work covers the many stages of a long journey that started with conceiving an innovative and challenging optical design, followed by the engineering and system engineering efforts to build it, to finally delivering an optimized and revolutionary cutting-edge facility.

astro-ph.IM↗
arXiv2026

Epitaxial Strain Activates Altermagnetic Spin-Splitting Torques in RuO2(100)

The altermagnetic nature of rutile RuO2 remains under active debate: bulk measurements indicate a nearly nonmagnetic ground state, whereas thin-film studies have reported symmetry-dependent transport signatures consistent with altermagnetism. Here, we provide experimental evidence that altermagnetic spin splitting in RuO2 is a strain-stabilized emergent state rather than an intrinsic bulk property. Angular-resolved spin-torque measurements reveal a symmetry-selected spin Hall response characteristic of altermagnetic spin splitting, which is strongest in the strained regime but progressively suppressed as the lattice relaxes toward the bulk limit. Complementary magnetic measurements further reveal enhanced coercivity and exchange-bias behavior exclusively in strained films, indicating the emergence of a strain-stabilized magnetic state. First-principles calculations reproduce the strain-dependent evolution of the Neel order and spin-split electronic structure, supporting the experimental observations. Together, these results establish altermagnetic spin splitting in RuO2 as a strain-stabilized emergent state and provide a unified explanation for the long-standing discrepancy between bulk and thin-film observations.

cond-mat.mtrl-sci↗
arXiv2026

The Origins of Stochasticity: Comprehensive Investigations on Uncertainty Quantification for Large Language Models

Recent advancements in Large Language Models (LLMs) have enabled sophisticated reasoning and content generation, yet their inherent stochasticity poses significant challenges for ensuring predictive credibility. While traditional uncertainty taxonomy paradigms, such as the dichotomy of aleatoric and epistemic uncertainties, provide conceptual foundations, they often fail to capture the multi-component and multi-stage nature of LLM generation and struggle to evaluate the effectiveness of various Uncertainty Quantification (UQ) methods. In this paper, we propose a granular uncertainty taxonomy that systematically attributes LLM uncertainty into input-level, parameter-level, token-level, and decoding-process sources. Correspondingly, we categorize existing UQ methods into Bayesian, ensemble, consensus-based, and single-pass approaches. Furthermore, we introduce a comprehensive evaluation framework covering diverse generation settings and metrics. We empirically evaluate 21 typical UQ methods across three prominent LLM families, including Qwen3, Llama 3.2, and DeepSeek-V3, on benchmarks such as TriviaQA, GSM8K, and HumanEval. Our experimental results demonstrate that (i) the effectiveness of UQ methods is sensitive to task types and generation settings; (ii) consensus-based methods, typed Deg and EigV, consistently outperform other UQ approaches; and (iii) larger model scales correlate with lower uncertainty estimates, suggesting an empirical scaling law for LLM uncertainty. This work bridges the gap between theoretical origins and practical deployment, providing a versatile diagnostic tool for systematically quantifying uncertainty in LLM applications.

cs.AI↗