arXiv subjects

Siyuan Cheng

Publications and source records attributed to Siyuan Cheng.

At least 19 recordsLinked to original sources

arXiv2026

Beyond Sparse Rewards: A New Benchmark and Structure-Aware Graph Alignment for Micro-Drama Understanding

Micro-dramas, characterized by ultra-short durations and hyper-dense storylines, pose unique challenges for video understanding that conventional benchmarks fail to address. To bridge this gap, we introduce M-Drama, the first large-scale bilingual benchmark for micro-drama comprehension, featuring over 35K instances across 9,138 clips. Furthermore, while reinforcement learning can enhance VLMs on complex narratives, existing reward metrics often suffer from sparse and superficial signals, failing to capture intricate character identities and temporal structures. We propose SAGA (Structure-Aware Graph Alignment), a novel graph-matching reward function that models narratives as heterogeneous graphs. SAGA computes dense, rigorous rewards via decoupled semantic triplet and structural temporal matching. Extensive experiments on Qwen3-VL-8B-Instruct demonstrate that SAGA outperforms existing baselines, delivering substantial improvements in open-ended accuracy and summary quality, while maintaining competitive out-of-domain generalization. Code is available at https://github.com/qyx1121/MDrama_SAGA.

cs.AI↗

arXiv2026

Low-Depth Random Unitaries without Ancillae

Random unitaries are fundamental to quantum information and many-body physics, with widespread applications ranging from quantum learning and metrology to device benchmarking. A central pursuit is to minimize the space and circuit depth required to generate them. However, existing methods for generating low-depth random unitaries rely heavily on an extensive number of ancillary qubits, imposing severe spatial overhead. In this work, we prove that random unitaries can be generated in optimal depth without ancillae. For multiplicative-error approximate $k$-designs on $n$ qubits, our circuits achieve a depth of $\widetilde{O} (k) (\log n)^{1/\delta}$ on $\delta$-dimensional architectures and $\widetilde{O}(k) \log \log n$ with all-to-all connectivity. Furthermore, by introducing a general exactification lemma, we lift our construction to optimal-depth exact $k$-designs, yielding an exponential resource reduction over state-of-the-art exact constructions. Our results minimize the space-time costs for a wide range of quantum protocols.

quant-ph↗

arXiv2026

Natural-Language Workflows Are Not Software Yet: Artifact-Driven Compilation for Reliable Agent Execution

Natural-language workflows offer a software-like interface for agents: domain experts can write reusable procedures, and agents can execute them as instructions. This promise is not yet reliable. Workflow descriptions often leave data dependencies implicit, so the executor must infer which prior results a step should use; agents can also fail to follow long or branching instructions under context pressure. We propose Artic, an artifact-driven workflow compiler that transforms a natural-language workflow into an artifact-driven workflow in which each step declares the artifacts it reads and writes, constraints gate produced artifacts, and explicit control transfers route execution. This representation exposes the enforcement burden placed on agent execution, allowing the compiler to identify steps that depend on too much state or contain difficult control logic and refine them through constrained optimization. To validate the LLM-assisted transformation, Artic decomposes faithfulness checking into local obligations and uses scenario-based dry runs to test whether compiled workflow regions conform to the source workflow. We evaluate Artic on 488 problem instances from 11 real-world domain workflows; it improves task resolve rate by 28 percentage points over the original text workflow. We also show that workflows compiled by Artic are 32 and 56 percentage points more consistent in cross-model and repeated-execution setups, respectively.

cs.SE↗

arXiv2026

No Cloning of Quantum Ensembles

Modern quantum physics now enables control of quantum systems at the level of individual trajectories, opening a new frontier that links quantum information theory, quantum many-body physics, and quantum thermodynamics, and uncovers novel non-equilibrium phenomena such as deep thermalization and measurement-induced entanglement. However, a central challenge remains: their characterization relies on measuring nonlinear properties of individual quantum states, a task tantamount to fine-grained cloning of a quantum ensemble. Here, the fundamental laws governing the cloning of quantum ensembles are investigated. First, a general no-cloning theorem for arbitrary ensembles is established from an information-theoretic perspective, even assuming multiple copies of the ensemble's purification. It is then shown that this barrier can be unexpectedly circumvented for physical ensembles generated by finite-time evolutions. Nevertheless, these tasks are proven to remain computationally intractable, even when the full circuit description of state preparation is known. This stands in sharp contrast to the conventional no-cloning theorem, which relies on the state being unknown. Together, these results establish new fundamental principles of quantum mechanics, reveal intrinsic trade-offs among sample complexity, computational complexity, and quantum measurements, and highlight the necessity of problem-specific strategies for probing measurement-induced quantum phenomena.

quant-ph↗

arXiv2026

Complexity-driven transitions in quantum observation

Observing the physical world is a foundational pursuit in science. In the quantum realm, however, observation necessitates a fundamental quantum-to-classical conversion: destructive measurements irreversibly project quantum states into classical data, inevitably incurring a loss of information. What physical principles govern this information loss, and how can we construct optimal measurements to maximize the readout? Here, we address these questions by establishing an intrinsic relationship between readout capability--quantified by the ratio of accessible classical Fisher information to the total quantum Fisher information (QFI), and measurement complexity--defined as the quantum circuit depth required prior to projection. Remarkably, we uncover a sudden emergence of observability: a sharp hidden-to-visible transition driven entirely by measurement complexity. We rigorously prove that below critical depth thresholds--$\Theta((\log n)^{1/\delta})$ for $\delta$-dimensional architectures and $\Theta(\log\log n)$ for all-to-all connectivity--readout capability decays exponentially with system size $n$, rendering the quantum information fundamentally inaccessible. Surprisingly, immediately above this threshold, the system enters a visible regime: we demonstrate that randomized measurements universally recover a constant fraction of the QFI using approximate unitary 3-designs, for which we explicitly develop optimal-depth circuit constructions tailored to finite-dimensional architectures. By unveiling the fundamental scaling laws and transitions that govern quantum observation, our results delineate definitive resource boundaries for quantum learning, state certification, and quantum metrology.

quant-ph↗

arXiv2026

Biorthogonal dynamical quantum phase transitions in non-Hermitian topological superconductors

Dynamical quantum phase transitions in non-Hermitian systems pose fundamental challenges due to the intrinsic biorthogonality of their eigenstates. In this work, we extend a biorthogonal framework to investigate dynamical quantum phase transitions in non-Hermitian topological superconductors. Taking the non-Hermitian Kitaev chain as a prototypical model, we construct an associated-state formalism and reformulate the Loschmidt rate function, dynamical topological order parameter, and dynamical Fisher zeros. Within this framework, we find that the critical times at which dynamical quantum phase transitions occur differ from those based on the conventional self-normal approaches. We further analyze momentum-resolved subsystems at critical momenta and demonstrate the robustness of the biorthogonal framework. Our work highlights the essential role of biorthogonality in nonequilibrium dynamics and establishes a consistent theoretical framework for dynamical quantum phase transitions in non-Hermitian topological superconductors.

quant-ph↗

arXiv2026

PRISM-MCTS: Learning from Reasoning Trajectories with Metacognitive Reflection

PRISM-MCTS: Learning from Reasoning Trajectories with Metacognitive Reflection Siyuan Cheng, Bozhong Tian, Yanchao Hao, Zheng Wei Published: 06 Apr 2026, Last Modified: 06 Apr 2026 ACL 2026 Findings Conference, Area Chairs, Reviewers, Publication Chairs, Authors Revisions BibTeX CC BY 4.0 Keywords: Efficient/Low-Resource Methods for NLP, Generation, Question Answering Abstract: The emergence of reasoning models, exemplified by OpenAI o1, signifies a transition from intuitive to deliberative cognition, effectively reorienting the scaling laws from pre-training paradigms toward test-time computation. While Monte Carlo Tree Search (MCTS) has shown promise in this domain, existing approaches typically treat each rollout as an isolated trajectory. This lack of information sharing leads to severe inefficiency and substantial computational redundancy, as the search process fails to leverage insights from prior explorations. To address these limitations, we propose PRISM-MCTS, a novel reasoning framework that draws inspiration from human parallel thinking and reflective processes. PRISM-MCTS integrates a Process Reward Model (PRM) with a dynamic shared memory, capturing both "Heuristics" and "Fallacies". By reinforcing successful strategies and pruning error-prone branches, PRISM-MCTS effectively achieves refinement. Furthermore, we develop a data-efficient training strategy for the PRM, achieving high-fidelity evaluation under a few-shot regime. Empirical evaluations across diverse reasoning benchmarks substantiate the efficacy of PRISM-MCTS. Notably, it halves the trajectory requirements on GPQA while surpassing MCTS-RAG and Search-o1, demonstrating that it scales inference by reasoning judiciously rather than exhaustively.

cs.AI↗

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

Entanglement signatures of quantum criticality in Floquet non-Hermitian topological systems

The entanglement entropy can be an effective diagnostic tool for probing topological phase transitions. In one-dimensional single particle systems, the periodic driving generates a variety of topological phases and edge modes. In this work, we investigate the topological phase transition of the one-dimensional Floquet Su-Schrieffer-Heeger model using entanglement entropy, and construct the phase diagram based on entanglement entropy. The entanglement entropy exhibits pronounced peaks and follows the logarithmic scaling law at the phase transition points, from which we extract the central charge $c=1$. We further investigate the entanglement spectrum to accurately distinguish the different topological phases. In addition, the coupling between zero and $\pi$ modes leads to characteristic splittings in the entanglement spectrum, signaling their hybridization under periodic driving. These results remain robust in non-Hermitian regimes and in the presence of next-nearest-neighbor hopping, demonstrating the reliability and universality of entanglement entropy as a diagnostic for topological phase transitions.

quant-ph↗

arXiv2025

Enhancing End-to-End Autonomous Driving with Risk Semantic Distillaion from VLM

The autonomous driving (AD) system has exhibited remarkable performance in complex driving scenarios. However, generalization is still a key limitation for the current system, which refers to the ability to handle unseen scenarios or unfamiliar sensor configurations.Related works have explored the use of Vision-Language Models (VLMs) to address few-shot or zero-shot tasks. While promising, these methods introduce a new challenge: the emergence of a hybrid AD system, where two distinct systems are used to plan a trajectory, leading to potential inconsistencies. Alternative research directions have explored Vision-Language-Action (VLA) frameworks that generate control actions from VLM directly. However, these end-to-end solutions demonstrate prohibitive computational demands. To overcome these challenges, we introduce Risk Semantic Distillation (RSD), a novel framework that leverages VLMs to enhance the training of End-to-End (E2E) AD backbones. By providing risk attention for key objects, RSD addresses the issue of generalization. Specifically, we introduce RiskHead, a plug-in module that distills causal risk estimates from Vision-Language Models into Bird's-Eye-View (BEV) features, yielding interpretable risk-attention maps.This approach allows BEV features to learn richer and more nuanced risk attention representations, which directly enhance the model's ability to handle spatial boundaries and risky objects.By focusing on risk attention, RSD aligns better with human-like driving behavior, which is essential to navigate in complex and dynamic environments. Our experiments on the Bench2Drive benchmark demonstrate the effectiveness of RSD in managing complex and unpredictable driving conditions. Due to the enhanced BEV representations enabled by RSD, we observed a significant improvement in both perception and planning capabilities.

cs.CV↗

arXiv2025

From Poisoned to Aware: Fostering Backdoor Self-Awareness in LLMs

Large Language Models (LLMs) can acquire deceptive behaviors through backdoor attacks, where the model executes prohibited actions whenever secret triggers appear in the input. Existing safety training methods largely fail to address this vulnerability, due to the inherent difficulty of uncovering hidden triggers implanted in the model. Motivated by recent findings on LLMs' situational awareness, we propose a novel post-training framework that cultivates self-awareness of backdoor risks and enables models to articulate implanted triggers even when they are absent from the prompt. At its core, our approach introduces an inversion-inspired reinforcement learning framework that encourages models to introspectively reason about their own behaviors and reverse-engineer the triggers responsible for misaligned outputs. Guided by curated reward signals, this process transforms a poisoned model into one capable of precisely identifying its implanted trigger. Surprisingly, we observe that such backdoor self-awareness emerges abruptly within a short training window, resembling a phase transition in capability. Building on this emergent property, we further present two complementary defense strategies for mitigating and detecting backdoor threats. Experiments on five backdoor attacks, compared against six baseline methods, demonstrate that our approach has strong potential to improve the robustness of LLMs against backdoor risks. The code is available at LLM Backdoor Self-Awareness.

cs.CR↗

arXiv2025

ASTRA: Autonomous Spatial-Temporal Red-teaming for AI Software Assistants

AI coding assistants like GitHub Copilot are rapidly transforming software development, but their safety remains deeply uncertain-especially in high-stakes domains like cybersecurity. Current red-teaming tools often rely on fixed benchmarks or unrealistic prompts, missing many real-world vulnerabilities. We present ASTRA, an automated agent system designed to systematically uncover safety flaws in AI-driven code generation and security guidance systems. ASTRA works in three stages: (1) it builds structured domain-specific knowledge graphs that model complex software tasks and known weaknesses; (2) it performs online vulnerability exploration of each target model by adaptively probing both its input space, i.e., the spatial exploration, and its reasoning processes, i.e., the temporal exploration, guided by the knowledge graphs; and (3) it generates high-quality violation-inducing cases to improve model alignment. Unlike prior methods, ASTRA focuses on realistic inputs-requests that developers might actually ask-and uses both offline abstraction guided domain modeling and online domain knowledge graph adaptation to surface corner-case vulnerabilities. Across two major evaluation domains, ASTRA finds 11-66% more issues than existing techniques and produces test cases that lead to 17% more effective alignment training, showing its practical value for building safer AI systems.

cs.CR↗

arXiv2025

SOFT: Selective Data Obfuscation for Protecting LLM Fine-tuning against Membership Inference Attacks

Large language models (LLMs) have achieved remarkable success and are widely adopted for diverse applications. However, fine-tuning these models often involves private or sensitive information, raising critical privacy concerns. In this work, we conduct the first comprehensive study evaluating the vulnerability of fine-tuned LLMs to membership inference attacks (MIAs). Our empirical analysis demonstrates that MIAs exploit the loss reduction during fine-tuning, making them highly effective in revealing membership information. These findings motivate the development of our defense. We propose SOFT (\textbf{S}elective data \textbf{O}bfuscation in LLM \textbf{F}ine-\textbf{T}uning), a novel defense technique that mitigates privacy leakage by leveraging influential data selection with an adjustable parameter to balance utility preservation and privacy protection. Our extensive experiments span six diverse domains and multiple LLM architectures and scales. Results show that SOFT effectively reduces privacy risks while maintaining competitive model performance, offering a practical and scalable solution to safeguard sensitive information in fine-tuned LLMs.

cs.CR↗

arXiv2025

ChineseHarm-Bench: A Chinese Harmful Content Detection Benchmark

Large language models (LLMs) have been increasingly applied to automated harmful content detection tasks, assisting moderators in identifying policy violations and improving the overall efficiency and accuracy of content review. However, existing resources for harmful content detection are predominantly focused on English, with Chinese datasets remaining scarce and often limited in scope. We present a comprehensive, professionally annotated benchmark for Chinese content harm detection, which covers six representative categories and is constructed entirely from real-world data. Our annotation process further yields a knowledge rule base that provides explicit expert knowledge to assist LLMs in Chinese harmful content detection. In addition, we propose a knowledge-augmented baseline that integrates both human-annotated knowledge rules and implicit knowledge from large language models, enabling smaller models to achieve performance comparable to state-of-the-art LLMs. Code and data are available at https://github.com/zjunlp/ChineseHarm-bench.

cs.CL↗

arXiv2025

CodeMirage: A Multi-Lingual Benchmark for Detecting AI-Generated and Paraphrased Source Code from Production-Level LLMs

Large language models (LLMs) have become integral to modern software development, producing vast amounts of AI-generated source code. While these models boost programming productivity, their misuse introduces critical risks, including code plagiarism, license violations, and the propagation of insecure programs. As a result, robust detection of AI-generated code is essential. To support the development of such detectors, a comprehensive benchmark that reflects real-world conditions is crucial. However, existing benchmarks fall short -- most cover only a limited set of programming languages and rely on less capable generative models. In this paper, we present CodeMirage, a comprehensive benchmark that addresses these limitations through three major advancements: (1) it spans ten widely used programming languages, (2) includes both original and paraphrased code samples, and (3) incorporates outputs from ten state-of-the-art production-level LLMs, including both reasoning and non-reasoning models from six major providers. Using CodeMirage, we evaluate ten representative detectors across four methodological paradigms under four realistic evaluation configurations, reporting results using three complementary metrics. Our analysis reveals nine key findings that uncover the strengths and weaknesses of current detectors, and identify critical challenges for future work. We believe CodeMirage offers a rigorous and practical testbed to advance the development of robust and generalizable AI-generated code detectors.

cs.SE↗

arXiv2025

Giant Orbital Torque-driven Picosecond Switching in Magnetic Tunnel Junctions

Orbital Hall effect was recently discovered as a novel pathway for driving magnetic moment. However, the integration of orbital Hall effect in magnetic memories suffers from low orbital-to-spin conversion efficiency and incompatibility with magnetic tunnel junctions. Here we demonstrate an orbital Hall effect-driven magnetic tunnel junction based on Ru/W bilayer, where the Ru layer possesses a strong orbital Hall conductivity and the {\alpha}-W layer features an orbital-to-spin conversion efficiency exceeding 90% because of the large orbit-spin diffusivity. By harnessing the giant orbital torque, we achieve a 28.7-picosecond switching and a five to eight-fold reduction in driving voltages over conventional spin-orbit torque magnetic memories. Our work bridges the critical gap between orbital effects and magnetic memory applications, significantly advancing the field of spintronics and orbitronics.

cond-mat.mtrl-sci↗

arXiv2025

CO-SPY: Combining Semantic and Pixel Features to Detect Synthetic Images by AI

With the rapid advancement of generative AI, it is now possible to synthesize high-quality images in a few seconds. Despite the power of these technologies, they raise significant concerns regarding misuse. Current efforts to distinguish between real and AI-generated images may lack generalization, being effective for only certain types of generative models and susceptible to post-processing techniques like JPEG compression. To overcome these limitations, we propose a novel framework, Co-Spy, that first enhances existing semantic features (e.g., the number of fingers in a hand) and artifact features (e.g., pixel value differences), and then adaptively integrates them to achieve more general and robust synthetic image detection. Additionally, we create Co-Spy-Bench, a comprehensive dataset comprising 5 real image datasets and 22 state-of-the-art generative models, including the latest models like FLUX. We also collect 50k synthetic images in the wild from the Internet to enable evaluation in a more practical setting. Our extensive evaluations demonstrate that our detector outperforms existing methods under identical training conditions, achieving an average accuracy improvement of approximately 11% to 34%. The code is available at https://github.com/Megum1/Co-Spy.

cs.CV↗

arXiv2025

CENSOR: Defense Against Gradient Inversion via Orthogonal Subspace Bayesian Sampling

Federated learning collaboratively trains a neural network on a global server, where each local client receives the current global model weights and sends back parameter updates (gradients) based on its local private data. The process of sending these model updates may leak client's private data information. Existing gradient inversion attacks can exploit this vulnerability to recover private training instances from a client's gradient vectors. Recently, researchers have proposed advanced gradient inversion techniques that existing defenses struggle to handle effectively. In this work, we present a novel defense tailored for large neural network models. Our defense capitalizes on the high dimensionality of the model parameters to perturb gradients within a subspace orthogonal to the original gradient. By leveraging cold posteriors over orthogonal subspaces, our defense implements a refined gradient update mechanism. This enables the selection of an optimal gradient that not only safeguards against gradient inversion attacks but also maintains model utility. We conduct comprehensive experiments across three different datasets and evaluate our defense against various state-of-the-art attacks and defenses. Code is available at https://censor-gradient.github.io.

cs.LG↗