Searcharxiv⌕ Search

arXiv subjects

Chenyu Wang

Publications and source records attributed to Chenyu Wang.

At least 55 records · Page 3Linked to original sources
arXiv2026

SPG: Sandwiched Policy Gradient for Masked Diffusion Language Models

Diffusion large language models (dLLMs) are emerging as an efficient alternative to autoregressive models due to their ability to decode multiple tokens in parallel. However, aligning dLLMs with human preferences or task-specific rewards via reinforcement learning (RL) is challenging because their intractable log-likelihood precludes the direct application of standard policy gradient methods. While prior work uses surrogates like the evidence lower bound (ELBO), these one-sided approximations can introduce significant policy gradient bias. To address this, we propose the Sandwiched Policy Gradient (SPG) that leverages both an upper and a lower bound of the true log-likelihood. Experiments show that SPG significantly outperforms baselines based on ELBO or one-step estimation. Specifically, SPG improves the accuracy over state-of-the-art RL methods for dLLMs by 3.6% in GSM8K, 2.6% in MATH500, 18.4% in Countdown and 27.0% in Sudoku.

cs.CL↗
arXiv2026

Enhancing Fine-Grained Spatial Grounding in 3D CT Report Generation via Discriminative Guidance

Vision--language models (VLMs) for radiology report generation (RRG) can produce long-form chest CT reports from volumetric scans and show strong potential to improve radiology workflow efficiency and consistency. However, existing methods face two key limitations: (i) training supervision is often coarse, aligning a whole CT volume with a full free-text report without explicit alignment for fine-grained attributes or pathology locations; and (ii) evaluation is typically holistic (lexical overlap, entity matching, or LLM-as-a-judge scores) and not diagnostic for spatial grounding. We propose \emph{Discriminative Cue-Prompting with Prompt Dropout (DCP-PD)}, a plug-and-play framework that distills fine-grained cues from free-text reports and uses them to guide report generation while mitigating shortcut reliance via prompt dropout. DCP-PD achieves state-of-the-art performance on CT-RATE, improving macro F1 from $=0.501$ to $0.603$ (20% relative), and substantially boosts out-of-distribution performance on Rad-ChestCT from F1 $=0.266$ to $0.503$ (89% relative). Finally, we introduce a hierarchical, location-aware question-set protocol (presence $\rightarrow$ laterality $\rightarrow$ lobe) to directly assess pathology-location grounding, showing that fine-grained spatial localization remains challenging even for models that score highly on current benchmarks.

cs.CV↗
arXiv2026

SPICE: Submodular Penalized Information-Conflict Selection for Efficient Large Language Model Training

Information-based data selection for instruction tuning is compelling: maximizing the log-determinant of the Fisher information yields a monotone submodular objective, enabling greedy algorithms to achieve a $(1-1/e)$ approximation under a cardinality budget. In practice, however, we identify alleviating gradient conflicts, misalignment between per-sample gradients, is a key factor that slows down the decay of marginal log-determinant information gains, thereby preventing significant loss of information. We formalize this via an $\varepsilon$-decomposition that quantifies the deviation from ideal submodularity as a function of conflict statistics, yielding data-dependent approximation factors that tighten as conflicts diminish. Guided by this analysis, we propose SPICE, a conflict-aware selector that maximizes information while penalizing misalignment, and that supports early stopping and proxy models for efficiency. Empirically, SPICE selects subsets with higher log-determinant information than original criteria, and these informational gains translate into performance improvements: across 8 benchmarks with LLaMA2-7B and Qwen2-7B, SPICE uses only 10% of the data, yet matches or exceeds 6 methods including full-data tuning. This achieves performance improvements with substantially lower training cost.

cs.LG↗
arXiv2026

Grafted Low-Leakage Si/AlN p-n Diodes Enabled by Fluorinated AlN Interface

Ultrawide-bandgap AlN is a promising material for next-generation power electronics; however, its practical implementation is hindered by unstable surface chemistry and the high activation energy of p-type dopants. In particular, high-temperature rapid thermal annealing (RTA), required for forming low-resistance contacts on n-type AlN, leads to the formation of thick and defective surface oxides that degrade heterojunction performance. In this work, we present an interface engineering approach based on fluorination-induced AlFx formation combined with SiNx passivation to suppress defect-assisted leakage in p-Si/n-AlN heterojunction diodes fabricated via semiconductor grafting. A low-damage pseudo-atomic layer etching process is employed to remove RTA-induced oxides and restore a near-stoichiometric AlN surface. Subsequent XeF2 treatment forms an ultrathin AlFx layer, which is stabilized by an atomic-layer-deposited SiNx capping layer prior to p-Si nanomembrane integration. Electrical measurements show that the engineered AlFx/SiNx interface reduces reverse leakage current by several orders of magnitude compared to untreated or oxide-removed AlN surfaces, while preserving forward conduction characteristics. Temperature-dependent analysis indicates strong suppression of Poole-Frenkel emission and a shift of leakage onset to higher reverse bias, ultimately limited by bulk AlN crystal quality. X-ray photoelectron spectroscopy and transmission electron microscopy confirm the formation of Al-F bonds, reduced Al-O content, and the presence of a thin interfacial SiOx/SiON layer. These results establish AlFx/SiNx passivation as an effective strategy for stabilizing AlN interfaces and enabling low-leakage ultrawide-bandgap heterojunction devices.

cond-mat.mtrl-sci↗
arXiv2026

Phonon-modulated Kerr nonlinearity in ultrathin 2H-MoTe2

Controlling nonequilibrium responses in optically driven quantum materials is essential for advancing applications in energy conversion, ultrafast electronics, and quantum computation. Nonlinear optical spectroscopy serves as a powerful tool to investigate ultrafast electron and phonon dynamics in these systems; however, conventional nonlinear approaches often require intense laser pulses (> 10 GW/cm2) and typically encounter a strong background. Here, we introduce a phase-sensitive nonlinear spectroscopic technique that operates at low laser powers (~ 10 kW/cm2, pulse energies ~ 10 pJ) and enables real-time monitoring and active control of coherent phonons in a few-layer (three to five) thick 2H-MoTe2. Upon excitation with ultrashort (~ 10 fs) pump pulses, we achieve displacive excitation of coherent phonons, which periodically modulate the Kerr nonlinearity of the material, leading to cross-phase modulation (XPM) of a delayed probe pulse. This phase modulation induces spectral broadening and oscillations in the center of mass (COM) of the probe spectrum in time, enabling the detection of subtle nonlinear optical responses in a background-free manner. The nonlinear response can be selectively amplified or attenuated by adjusting the strength of the pump pulse, which controls the distribution of photoexcited carriers in the electronic bands. By combining two-color nondegenerate pump-probe measurements and time-dependent density-functional theory (TDDFT) calculations, we directly resolve the coupled nonequilibrium electronic and phonon dynamics. A dual-pump pulse scheme enables precise control of phonon oscillations, allowing selective activation or suppression of specific phonon modes and correspondingly the modulation of the Kerr nonlinearity.

physics.optics↗
arXiv2026

virtCCA: Virtualized Arm Confidential Compute Architecture with TrustZone

Arm introduced the Confidential Compute Architecture (CCA) in the upcoming Armv9-A architecture, enabling the support of confidential virtual machines (CVMs) in a separate world called the Realm world, providing protection from untrusted normal world. While CCA offers a promising future for confidential computing, the widespread commercial CCA hardware is not available in the near future. To fill this gap, we present virtCCA, an architecture that facilitates virtualized CCA using TrustZone, a mature hardware feature on existing Arm platforms. Notably, virtCCA can be implemented on platforms equipped with the Secure EL2 (S-EL2) extension from ARMv8.4 onwards, as well as on earlier platforms that lack S-EL2 support. virtCCA provides strong compatibility with the CCA specifications at the API level. We developed the entire CCA software and firmware stack on top of virtCCA, including the enhancements to the normal world's KVM to support CVMs, and the TrustZone Management Monitor (TMM) that enforces isolation among CVMs and provides CVM lifecycle management. We have implemented virtCCA on real Arm servers, both with and without S-EL2 support. Our evaluation on micro-benchmarks and macrobenchmarks demonstrates that the overhead of running CVMs is acceptable compared to running normal-world VMs. Specifically, in a set of real-world workloads, the overhead of virtCCA-SEL2 is less than 29.7% for I/O intensive workloads, while virtCCA-EL3 outperforms the baseline in most cases.

cs.CR↗
arXiv2026

Trojans in Artificial Intelligence (TrojAI) Final Report

The Intelligence Advanced Research Projects Activity (IARPA) launched the TrojAI program to confront an emerging vulnerability in modern artificial intelligence: the threat of AI Trojans. These AI trojans are malicious, hidden backdoors intentionally embedded within an AI model that can cause a system to fail in unexpected ways, or allow a malicious actor to hijack the AI model at will. This multi-year initiative helped to map out the complex nature of the threat, pioneered foundational detection methods, and identified unsolved challenges that require ongoing attention by the burgeoning AI security field. This report synthesizes the program's key findings, including methodologies for detection through weight analysis and trigger inversion, as well as approaches for mitigating Trojan risks in deployed models. Comprehensive test and evaluation results highlight detector performance, sensitivity, and the prevalence of "natural" Trojans. The report concludes with lessons learned and recommendations for advancing AI security research.

cs.CR↗
arXiv2026

Slm-mux: Orchestrating small language models for reasoning

With the rapid development of language models, the number of small language models (SLMs) has grown significantly. Although they do not achieve state-of-the-art accuracy, they are more efficient and often excel at specific tasks. This raises a natural question: can multiple SLMs be orchestrated into a system where each contributes effectively, achieving higher accuracy than any individual model? Existing orchestration methods have primarily targeted frontier models (e.g., GPT-4) and perform suboptimally when applied to SLMs. To address this gap, we propose a three-stage approach for orchestrating SLMs. First, we introduce SLM-MUX, a multi-model architecture that effectively coordinates multiple SLMs. Building on this, we develop two optimization strategies: (i) a model selection search that identifies the most complementary SLMs from a given pool, and (ii) test-time scaling tailored to SLM-MUX. Our approach delivers strong results: Compared to existing orchestration methods, our approach achieves up to 13.4% improvement on MATH, 8.8% on GPQA, and 7.0% on GSM8K. With just two SLMs, SLM-MUX outperforms Qwen 2.5 72B on GPQA and GSM8K, and matches its performance on MATH. We further provide theoretical analyses to substantiate the advantages of our method. Additional experiments show that the core principle of SLM-MUX extends to open-ended generation tasks (e.g., HumanEval) and benefits other model classes, including frontier LLMs and domain-specific fine-tuned SLMs. In summary, we demonstrate that SLMs can be effectively orchestrated into more accurate and efficient systems through the proposed approach. The project page is available at https://slm-mux.github.io/.

cs.CL↗
arXiv2026

GenAI for Systems: Recurring Challenges and Design Principles from Software to Silicon

Generative AI is reshaping how computing systems are designed, optimized, and built, yet research remains fragmented across software, architecture, and chip design communities. This paper takes a cross-stack perspective, examining how generative models are being applied from code generation and distributed runtimes through hardware design space exploration to RTL synthesis, physical layout, and verification. Rather than reviewing each layer in isolation, we analyze how the same structural difficulties and effective responses recur across the stack. Our central finding is one of convergence. Despite the diversity of domains and tools, the field keeps encountering five recurring challenges (the feedback loop crisis, the tacit knowledge problem, trust and validation, co-design across boundaries, and the shift from determinism to dynamism) and keeps arriving at five design principles that independently emerge as effective responses (embracing hybrid approaches, designing for continuous feedback, separating concerns by role, matching methods to problem structure, and building on decades of systems knowledge). We organize these into a challenge--principle map that serves as a diagnostic and design aid, showing which principles have proven effective for which challenges across layers. Through concrete cross-stack examples, we show how systems navigate this map as they mature, and argue that the field needs shared engineering methodology, including common vocabularies, cross-layer benchmarks, and systematic design practices, so that progress compounds across communities rather than being rediscovered in each one. Our analysis covers more than 275 papers spanning eleven application areas across three layers of the computing stack, and distills open research questions that become visible only from a cross-layer vantage point.

cs.SE↗
arXiv2026

Kirin: Improving ANN efficiency with SNN Hybridization

Artificial neural networks (ANNs), particularly large language models (LLMs), demonstrate powerful inference capabilities but consume substantial energy. Conversely, spiking neural networks (SNNs) exhibit exceptional energy efficiency due to their binary and event-driven characteristics, thus motivating the study of ANN-to-SNN conversion. In this process, quantization plays a pivotal role, mapping LLMs' floating-point parameters to discrete SNN parameters via the temporal dimension of the time window. However, several challenges remain in the conversion process: (i) converting high bit-width quantization values into binary spikes requires longer time windows, increasing system latency; and (ii) the inherent trade-off between the information loss of single-spike schemes and the energy costs of multi-spike ones in SNN. To address these challenges, we propose Kirin, a integer and spike hybrid based SNN to achieve accuracy lossless ANN-to-SNN conversion with time and energy efficiency. Specifically, we first propose a Spike Matrix Hybridization strategy that encoding low bit-width parameters that leading to small time window size into binary spikes while preserving the rest in integer format, thereby reducing the overall latency of SNN execution. Second, we introduce a silence threshold mechanism to regulate the timing of single-spike firing, ensuring the output is mathematically equivalent to the LLM's output and preserves accuracy. Experimental results demonstrate that Kirin, under a W4A4\&8 quantization setting, achieves near-FP16 accuracy while reducing energy consumption by up to 84.66\% and shortening time steps by 93.75\%.

cs.LG↗
arXiv2026

VLM-UQBench: A Benchmark for Modality-Specific and Cross-Modality Uncertainties in Vision Language Models

Uncertainty quantification (UQ) is vital for ensuring that vision-language models (VLMs) behave safely and reliably. A central challenge is to localize uncertainty to its source, determining whether it arises from the image, the text, or misalignment between the two. We introduce VLM-UQBench, a benchmark for modality-specific and cross-modal data uncertainty in VLMs, It consists of 600 real-world samples drawn from the VizWiz dataset, curated into clean, image-, text-, and cross-modal uncertainty subsets, and a scalable perturbation pipeline with 8 visual, 5 textual, and 3 cross-modal perturbations. We further propose two simple metrics that quantify the sensitivity of UQ scores to these perturbations and their correlation with hallucinations, and use them to evaluate a range of UQ methods across four VLMs and three datasets. Empirically, we find that: (i) existing UQ methods exhibit strong modality-specific specialization and substantial dependence on the underlying VLM, (ii) modality-specific uncertainty frequently co-occurs with hallucinations while current UQ scores provide only weak and inconsistent risk signals, and (iii) although UQ methods can rival reasoning-based chain-of-thought baselines on overt, group-level ambiguity, they largely fail to detect the subtle, instance-level ambiguity introduced by our perturbation pipeline. These results highlight a significant gap between current UQ practices and the fine-grained, modality-aware uncertainty required for reliable VLM deployment.

cs.CV↗
arXiv2026

Generalist Foundation Models from a Multimodal Dataset for 3D Computed Tomography

Advancements in medical imaging AI, particularly in 3D imaging, have been limited due to the scarcity of comprehensive datasets. We introduce CT-RATE, a public dataset that pairs 3D medical images with corresponding textual reports. CT-RATE comprises 25,692 non-contrast 3D chest CT scans from 21,304 unique patients. Each scan is accompanied by its corresponding radiology report. Leveraging CT-RATE, we develop CT-CLIP, a CT-focused contrastive language-image pretraining framework designed for broad applications without the need for task-specific training. We demonstrate how CT-CLIP can be used in multi-abnormality detection and case retrieval, and outperforms state-of-the-art fully supervised models across all key metrics. By combining CT-CLIP's vision encoder with a pretrained large language model, we create CT-CHAT, a vision-language foundational chat model for 3D chest CT volumes. Finetuned on over 2.7 million question-answer pairs derived from the CT-RATE dataset, CT-CHAT underscores the necessity for specialized methods in 3D medical imaging. Collectively, the open-source release of CT-RATE, CT-CLIP, and CT-CHAT not only addresses critical challenges in 3D medical imaging but also lays the groundwork for future innovations in medical AI and improved patient care.

cs.CV↗
arXiv2026

Ribbons from Independence Structure: Hypercontractivity, $Φ$-Mutual Information, and Matrix $Φ$-Entropy

We study the hypercontractivity ribbon and the $Φ$-ribbon for joint distributions that obey a given independence structure, obtaining tight bounds in some basic regimes. For general independence structures, modeled as a hypergraph whose hyperedges specify mutually independent subcollections of random variables, we provide an explicit inner bound on the $Φ$-ribbon described by a simple convex hull of incidence vectors. We also provide a new multipartite generalization version and a $Φ$-mutual information analogue of the Zhang--Yeung inequality, which implies nontrivial points in the hypercontractivity ribbon and the $Φ$-ribbon respectively. Finally, we propose the matrix $Φ$-ribbon based on matrix $Φ$-entropy and establish the tensorization and data processing properties, together with the calculation of an exact matrix SDPI constant for the doubly symmetric binary source.

cs.IT↗
arXiv2025

Extreme Nanoconfinement Reshapes the Self-Dissociation of Water

Water's ability to self-dissociate into H$_3$O$^+$ and OH$^-$ ions is central to acid-base chemistry and bioenergetics. Recent experimental advances have enabled the confinement of water down to the nanometre scale, even to the single-molecule limit, yet how this process is altered at the extreme nanoconfinement remains unclear. Using \emph{ab-initio} calculations and enhanced-sampling machine-learning potential molecular dynamics, we show that monolayer-confined water exhibits a markedly lower barrier to auto-dissociation than bulk water. Confinement restructures both intramolecular bonding and the intermolecular hydrogen-bond network, while enforcing quasi-2D dipolar correlations that amplify dielectric fluctuations. Our results imply that two-dimensional confined water could act as a \emph{superdielectric} medium and may exhibit \emph{superionic} behavior, as observed in recent experiments. These findings reveal confinement as a powerful route to enhanced proton activity, shedding light on geochemical niches, biomolecular environments, and nanofluidic systems where water's chemistry is fundamentally reshaped.

cond-mat.mes-hall↗
arXiv2025

Kling-Omni Technical Report

We present Kling-Omni, a generalist generative framework designed to synthesize high-fidelity videos directly from multimodal visual language inputs. Adopting an end-to-end perspective, Kling-Omni bridges the functional separation among diverse video generation, editing, and intelligent reasoning tasks, integrating them into a holistic system. Unlike disjointed pipeline approaches, Kling-Omni supports a diverse range of user inputs, including text instructions, reference images, and video contexts, processing them into a unified multimodal representation to deliver cinematic-quality and highly-intelligent video content creation. To support these capabilities, we constructed a comprehensive data system that serves as the foundation for multimodal video creation. The framework is further empowered by efficient large-scale pre-training strategies and infrastructure optimizations for inference. Comprehensive evaluations reveal that Kling-Omni demonstrates exceptional capabilities in in-context generation, reasoning-based editing, and multimodal instruction following. Moving beyond a content creation tool, we believe Kling-Omni is a pivotal advancement toward multimodal world simulators capable of perceiving, reasoning, generating and interacting with the dynamic and complex worlds.

cs.CV↗
arXiv2025

Preliminary Demonstration of Diamond-GaN pn Diodes via Grafting

Ultrawide bandgap (UWBG) semiconductors exhibit exceptional electrical and thermal properties, offering strong potential for high power and high frequency electronics. However, efficient doping in UWBG materials is typically limited to either n type or p type, constraining their application to unipolar devices. The realization of pn junctions through heterogeneous integration of complementary UWBG or WBG semiconductors is hindered by lattice mismatch and thermal expansion differences. Here, we report the preliminary demonstration of diamond GaN heterojunction pn diodes fabricated via grafting. A single crystalline p plus diamond nanomembrane was integrated onto an epitaxially grown c plane n plus GaN substrate with an ultrathin ALD Al2O3 interlayer. The resulting diodes exhibit an ideality factor of 1.55 and a rectification ratio of over 1e4. Structural and interfacial properties were examined by AFM, XRD, Raman, and STEM, providing critical insights to guide further optimization of diamond GaN pn heterojunction devices.

cond-mat.mtrl-sci↗
arXiv2025

Energy-Efficient and Dequantization-Free Q-LLMs: A Spiking Neural Network Approach to Salient Value Mitigation

In the era of large language models (LLMs), weight-activation quantization helps fit models on edge device by reducing memory and compute bit-widths. However, three challenges persist for energy constrained hardware: (1) even after quantization, multiply-accumulate (MAC) operations remain unavoidable and continue to dominate energy consumption; (2) dequantization (or per-tensor/channel rescaling) introduces extra arithmetic and data movement, increasing latency and energy; (3) uniform parameters bit widths clip salient values-while intra-channel mixed precision is generally impractical on current matrix hardware and memory. In contrast, brain-inspired Spiking Neural Networks (SNNs), owing to their binary spike-based information representation and the Integrate-and-Fire (IF) paradigm, naturally support mixed-precision storage and energy-efficient computation by replacing complex MACs with temporal Accumulate (ACCs). Motivated by this property, we propose SpikeQuant, which selectively applies mixed-precision quantization to activations with salient values and re-encodes them into binary spike counts, thereby enabling dynamic mixed storage of different bitwidths. Furthermore, by embedding the quantization scale into the threshold of the IF mechanism, our approach performs energy-efficient linear transformations on weights and activations while avoiding explicit dequantization. Experimental results demonstrate that SpikeQuant consistently achieves near-FP16 perplexity under W4A4 quantization while reducing energy cost by up to 4.6 times compared to existing methods, highlighting its effectiveness for accurate and energy-efficient LLM deployment.

cs.LG↗
arXiv2025

A Common Synchrotron Origin for Prompt Gamma-Ray and Soft X-Ray Emission in GRBs: Evidence from Joint Spectral Analysis

The recent launches of the Einstein Probe (EP) and the Space Variable Objects Monitor (SVOM) mission have led to the detection of a growing number of long GRBs with significant, early soft X-ray flux during their gamma-ray emission, prompting the question of whether their multi-band prompt emission shares a common origin in region and mechanism. To address this, we utilize the 20-year Swift archival data, which provides a substantial sample of joint soft X-ray and gamma-ray observations, enabling a systematic joint spectral study. We resolve 110 temporal pulses from 46 GRBs and find that a single power-law model with a low-energy break or cutoff adequately describes the prompt spectra from 150 keV down to 0.5 keV. More than half of the sample pulses require a break around a few keV, with average spectral indices $\langle α_1 \rangle = -0.88$ and $\langle α_2 \rangle = -1.46$ consistent with synchrotron radiation in a marginally fast-cooling regime. The observed spectral evolution and the distribution of indices support a single-emission-region origin, where the varying spectral shapes are largely governed by the evolution of the synchrotron cooling frequency $ν_c$ and the effect of finite emission width. The observed differences in the temporal behavior between X-ray and gamma-ray light curves can be naturally explained by this spectral evolution across the broad band.

astro-ph.HE↗