arXiv subjects

Qian Chen

Publications and source records attributed to Qian Chen.

At least 19 recordsLinked to original sources

arXiv2026

ConsensusBench: Benchmark of Consensus Nodes for LLM Reasoning via Outcome Reward Densifying

Reinforcement learning (RL) has become one of the primary paradigms for reasoning enhancement of large language models (LLMs). In particular, Group Relative Policy Optimization (GRPO) and related algorithms have demonstrated strong performance with outcome-level rewards. However, these methods depend solely on the final answer, without feedback regarding which intermediate steps contribute to success or failure. As task complexity and reasoning trajectory length increase, such sparse final-answer rewards become increasingly insufficient. To address this limitation, we introduce ConsensusBench, a novel dataset designed to provide rule-based process-level signals. We posit that a correct final answer relies on a small set of intermediate conclusions throughout the reasoning process, which can be seen as a verifiable sub-outcome. We identify these sub-outcomes by filtering correct trajectories from N rollouts and clustering semantically equivalent intermediate statements. We call these clustered statements as Consensus Nodes. By integrating a rule-based process reward derived from these nodes into GRPO-style algorithms, we develop a new reinforcement learning signal named ConsensusPR. It directly reduces the reward sparsity of outcome reward across long reasoning trajectories. To facilitate systematic process-level evaluation, we introduce three metrics to our benchmark: Final Answer Accuracy (Acc), Node Coverage Rate (NCR), and Tokens per Node (TPN). Experiments across AIME 2024, AIME 2025, GSM8K, MATH-500, and our ConsensusBench demonstrate that the proposed method consistently surpasses GRPO-style approaches, highlighting the practical value of consensus nodes in guiding reasoning.

cs.CL↗

arXiv2026

Most properties are undecidable even in $\mathop{\mathsf{NExt}} \mathsf{Grz}_t$

We investigate decidability of properties in the lattice $\mathop{\mathsf{NExt}} \mathsf{Grz}_t$ of extensions of the Grzegorczyk tense logic $\mathsf{Grz}_t$ and the lattice $\mathop{\mathsf{NExt}} \mathsf{S4}_t$ of reflexive and transitive tense logics, with applications to the lattice $\mathop{\mathsf{Ext}} \mathsf{biIPC}$ of bi-superintuitionistic logics. We prove that a broad class of properties is undecidable in $\mathop{\mathsf{NExt}} \mathsf{Grz}_t$, including tabularity, Kripke completeness, the finite model property, and decidability, which also yields their undecidability in $\mathop{\mathsf{NExt}} \mathsf{S4}_t$. We also construct infinitely many tabular extensions of $\mathsf{Grz}_t$ (and thus of $\mathsf{S4}_t$) whose coincidence problems are undecidable, while presenting one tabular extension of $\mathsf{Grz}_t$ and infinitely many ones of $\mathsf{S4}_t$ with a decidable coincidence problem. As a consequence, we obtain that the finite model property and tabularity are undecidable in $\mathop{\mathsf{Ext}} \mathsf{biIPC}$, and that there are infinitely many tabular extensions of $\mathsf{biIPC}$ whose coincidence problems are undecidable. These results clarify some similarities and differences between $\mathop{\mathsf{NExt}} \mathsf{Grz}_t$ and $\mathop{\mathsf{NExt}} \mathsf{Grz}$, $\mathop{\mathsf{NExt}} \mathsf{S4}_t$ and $\mathop{\mathsf{NExt}} \mathsf{S4}$, as well as $\mathop{\mathsf{Ext}} \mathsf{biIPC}$ and $\mathop{\mathsf{Ext}} \mathsf{IPC}$. The proofs adapt Chagrov's method of reducing from an undecidable problem for Minsky machines. We isolate and explicitly formulate the method of good valuations, a recurring technique underlying several proofs in the literature that use large frames, making it available for further applications.

math.LO↗

arXiv2026

OASIS: Optimizing Attacker Sequences for Hard-Label Black-Box Text Attacks

Different attack methods follow different search trajectories, they succeed on different subsets of samples, whereas existing hard-label black-box text attacks mainly focus on improving individual attackers or manually combining them. We present OASIS, a method for optimizing attacker sequences in hard-label black-box text attacks. OASIS first performs a one-time bi-objective attack chain search over candidate sequences to balance attack success rate and perturbation, and then reuses the selected fixed global chain during attack chain execution. Experiments across multiple datasets, victim models, and large language models show that OASIS consistently outperforms strong standalone baselines and simple manually constructed chains. These results suggest that attacker composition is not merely an implementation choice, but a practical optimization target for improving hard-label black-box text attacks.

cs.CR↗

arXiv2026

Short Horizons and Sparse Concepts: a Mathematical View of the Readout in the J-lens

The Jacobian lens (J-lens) has been proposed as a way to read verbalizable representations from language models. However, its principle and meaning lack a detailed and theoretical discussion. We provide a mathematical view of this interpretation and of its assumed causal structure. Besides treating the J-lens as a heuristic probe, we further regard it as a first-order causal transfer operator from intermediate activations to expected future readouts. We study the Jacobian matrix as the optimal local linear approximation of the downstream mapping, analyze its global approximation behavior and bias, and identify its mathematical meaning as an expectation over anticipated future readouts. Further analysis of the Jacobian energy distribution reveals that its causal geometry is highly sparse. The energy decays with depth, concentrates in an extremely small proportion, and decomposes into diagonal pathways and specific critical positions. This decomposition further resolves the expectation of the J-lens over future outputs into short-horizon and sparse concept predictions, providing a more intuitive attribution and explanation for the ability of the J-lens to visualize concepts during the thinking process. Based on the theory, we propose a simple but effective improvement strategy and decoupling method for the J-lens, which significantly enhances the ability of the J-lens to read out correct intermediate concepts.

cs.CL↗

arXiv2026

StreamFlow: Dynamic Memory Flows for Streaming Video Understanding

Streaming video understanding requires multimodal large language models (MLLMs) to preserve relevant evidence from continuously evolving streams under strict causality and bounded memory. Yet existing paradigms remain limited: model-based methods require intrusive backbone updates, while memory-based methods expend substantial visual-encoding computation on temporally redundant content and rely on rigid access to visual history. To address these limitations, we introduce StreamFlow, an efficient visual memory framework that enables dynamic, on-demand access to historical visual information. StreamFlow combines a lightweight, dynamics-aware mid-term memory that filters temporal redundancy before visual encoding with a latent long-term memory that consolidates historical video content into visual latents accessible to subsequent reasoning. During generation, an attention-guided retrieval mechanism injects relevant visual latents when the model's reliance on visual evidence weakens. StreamFlow achieves state-of-the-art streaming video understanding performance, reaching 67.73% overall accuracy on StreamingBench, while also delivering strong performance on offline long-video benchmarks. Relative to the vanilla setting, it improves the visual attention score (VAS) by 59.1% while reducing end-to-end latency and peak memory by 50.4% and 21.1%, respectively, enabling more visually grounded and efficient reasoning.

cs.CV↗

arXiv2026

GWM-VLA: Geometry-Aware Latent World Modeling for Vision-Language-Action Learning

Vision-Language-Action (VLA) models achieve strong robotic manipulation performance but often degrade under visual and environmental shifts. Latent world modeling offers a promising approach to improving robustness, yet existing methods commonly encode camera views independently and predict holistic scene dynamics without explicitly modeling their geometric relationships. We propose GWM-VLA, a geometry-aware latent world modeling framework for VLA learning. GWM-VLA combines geometry-aware multi-view state encoding, global context-conditioned target-view prediction, and shared latent-action representations grounded by robot-action supervision. Specifically, VGGT-$\Omega$ jointly aggregates multi-view observations at each timestep to construct geometry-aware multi-view states. The latent world model predicts the next-step patch tokens of a selected target view using patch and register tokens obtained after multi-view aggregation, thereby retaining multi-view geometric information without predicting the complete multi-view state. We use the wrist view as the target in our experiments, placing greater emphasis on end-effector motion and local gripper-object interactions. Finally, the shared latent-action representations condition both the latent world model and the flow-matching action head, allowing latent-prediction supervision and ground-truth robot-action supervision to jointly shape the same latent-action representations. Experiments across both simulation and real-world environments demonstrate the effectiveness and robustness of GWM-VLA.

cs.RO↗

arXiv2026

BALANCE: Hybrid Autoregressive-Speculative LLM Inference in Wireless Edge Networks

Edge inference is a promising paradigm to provide large language model (LLM) inference services in next-generation mobile networks. LLM inference mainly relies on two approaches: Autoregressive decoding (AD) generates output tokens sequentially, resulting in long latency; Speculative decoding (SD) accelerates inference by using a small language model (SLM) to generate multiple draft tokens for LLM verification, but incurs extra memory costs. Due to this latency-memory tradeoff, neither approach alone can efficiently serve users with heterogeneous demands under limited edge computing resources. To address this challenge, we propose a hybrid autoregressive-speculative inference (BALANCE) framework for edge LLM inference. In BALANCE, an edge server hosts both an SLM and an LLM, assigns each user to AD or SD, and performs the two modes simultaneously. To maximize the number of served users, we formulate a task throughput maximization problem to jointly determine user scheduling and computing resource allocation between AD and SD under user latency requirements and server memory constraints. Since the problem is NP-hard, we develop a polynomial-time algorithm that transforms the original problem into two sub-problems and obtains a sub-optimal solution with a constant approximation guarantee. Experiments demonstrate that BALANCE consistently outperforms conventional AD and SD and significantly improves task throughput.

cs.NI↗

arXiv2026

GROW: Group-Relative Advantage-Weighted On-Policy Reinforcement Learning of Autoregressive-Diffusion Text-to-Speech model

Reinforcement learning for flow-matching text-to-speech is complicated by deterministic ODE sampling: trajectory-level policy-gradient methods typically convert the ODE into an SDE and track per-step likelihood ratios, introducing stochastic perturbations and substantial overhead. We propose GROW, a group-relative advantage-weighted on-policy RL method that acts directly on the standard flow-matching objective. For each prompt, GROW samples a group of on-policy utterances, separately standardizes intelligibility and speaker-similarity rewards within the group, and combines them to reweight flow-matching regression. A Wasserstein-2 velocity penalty anchors the updated model to a frozen pretrained reference. A group-mean reward baseline is introduced to convert reward weighting into advantage weighting. For strong pretrained TTS models with concentrated rewards, positive exponential weighting is dominated by reward-agnostic self-imitation, whereas a zero-mean signed advantage preserves effective within-group credit assignment. Instantiated on DiTAR and evaluated on LibriSpeech and Seed-TTS EN/ZH, GROW reduces average WER from 2.016 to 1.558 and raises speaker similarity from 0.676 to 0.715 while keeping UTMOS. With 10-NFE training rollouts and 32-NFE evaluation, GROW retains comparable performance while training 2.9x faster than 32-NFE DiTAR-GRPO. We will open-source complete GROW codes, faithful DiTAR reproduction, and all model checkpoints.

eess.AS↗

arXiv2026

Enabling Proactive Spoken Turns via a Generalized Style-Aware Full-Duplex Framework

Compared with half-duplex dialogue systems where the system waits for user turn completion before it responds, natural full-duplex dialogue systems require agents to act proactively in real time, including timely interruptions and backchannels. This creates a key challenge: improving turn timing without sacrificing response quality. To address limitations in realistic proactive turn-taking, we build a generalized style-aware full-duplex framework with three key components. Firstly, we propose LPS-TC, a Lightweight Proactive Speech Turn Controller for plug-and-play integration. It features a fine-grained action space covering both reactive and proactive turn behaviors, enabling half-duplex models with full-duplex capabilities and enhancing existing full-duplex models with superior timing control. Secondly, we construct WildTurn, a large-scale, real-world English dataset containing approximately 2,981 hours of filtered multi-turn stereo conversations from face-to-face and telephone conversations, annotated with five turn-taking and five backchanneling styles. Trained on WildTurn, LPS-TC exhibits rich spoken dynamics that are not captured by existing static full-duplex benchmarks. Thirdly, we introduce a two-tier evaluation scheme that assesses both chunk-level timing precision and turn-level interaction quality under realistic streaming constraints. Our experiments, integrating LPS-TC with half-duplex models like Qwen2.5-Omni and full-duplex models like Freeze-Omni, showcase its superior performance in timing appropriateness and response quality. Our framework also demonstrates fine-grained style controllability and strong generalizability, enabling more natural and human-like spoken interactions.

cs.CL↗

arXiv2026

Capturing Token Tendencies for Training-Free Token Pruning in Multimodal Large Language Models

While visual token pruning is essential for efficient Multimodal Large Language Models (MLLMs), existing training-free methods suffer from a critical limitation: they rely on static, instantaneous heuristics to perform irreversible filtering. This approach ignores the hierarchical nature of MLLMs, where token importance often evolves dynamically rather than remaining fixed across layers. Consequently, tokens essential for deep-layer reasoning are often prematurely discarded by shallow-layer estimates. To address this, we propose Trend-aware Pruning, a novel framework that elevates pruning from a local snapshot decision to a temporal trajectory modeling problem. Instead of relying on isolated scores, our method captures the momentum of attention flow. This enables a dynamic rectification mechanism that selectively reactivates "late-blooming" tokens, those initially undervalued but exhibiting rising semantic importance, thereby preventing the loss of critical visual cues. Extensive experiments demonstrate that our approach achieves a superior efficiency-performance trade-off across diverse multimodal tasks. Notably, it reduces visual tokens by over 77.8%, retaining only approximately 23 tokens in the final layer while maintaining competitive performance, offering a robust and reversible solution for high-efficiency multimodal inference.

cs.CV↗

arXiv2026

Neural variational framework for random Young-diagram limit shapes

We develop a structure-preserving neural variational framework for random Young-diagram ensembles, with representations adapted to the structure and scaling of each measure. The method is validated on the Plancherel, uniform, minimal-difference, and fixed-$q$ $q$-Plancherel ensembles, using known asymptotic profiles only for post-training comparison. We then study a quartically deformed hook-length ensemble without assuming an analytical saddle shape. Large-$n$ neural profiles are compared with finite-size MAP profiles obtained from exact-action searches and with mean profiles obtained from corner-transfer Metropolis--Hastings sampling. Increasing the deformation suppresses the leading rows and broadens the support, while the neural, discrete, and sampled mean profiles agree at the percent level. These results provide numerical evidence for a deformation-dependent macroscopic saddle family.

cond-mat.stat-mech↗

arXiv2026

Spontaneous Scalarization of Brane Black Holes: Quantum-Enhanced Tachyonic Instabilities

We investigate the spontaneous scalarization of brane-localized charged black holes, focusing on the role of quantum-enhanced tachyonic instabilities. By solving the coupled bulk-brane equations numerically within the Einstein-DeTurck formulation, we construct fully backreacted static black hole solutions and map the system from the semi-classical limit to the strongly coupled quantum-dominated regime. We demonstrate that the holographic quantum effects, parameterized by $\kappa$, significantly modify the background geometry. Stability analysis, conducted via both effective potential diagnostics and quasinormal mode calculations, confirms that this geometric deformation deepens the negative well of the effective potential, thereby triggering a tachyonic instability. Integrating these results, we establish the global phase diagram of spontaneous scalarization in the $T_h/\mu-\kappa$ parameter space. The results reveal that these quantum-enhanced instabilities promote scalarization by substantially raising the critical temperature, providing a clear quantitative signature of quantum effects on black hole hair formation.

hep-th↗

arXiv2026

Efficient Chain-of-Modality Reasoning via Progressive Compression for Spoken Language Models

Spoken language models (SLMs) enable natural human-computer interaction, but their reasoning ability still lags behind that of text-based large language models, especially on spoken mathematical question answering tasks. One important reason is that SLMs reason over purely verbalized mathematical expressions, which are harder to interpret than symbolic text. However, directly transferring text-based reasoning to SLMs is nontrivial due to architectural constraints and the additional computational requirements. To address this challenge, we propose Efficient Chain-of-Modality Reasoning (ECoM Reasoning), the first framework to introduce compressed reasoning into SLMs. By compressing the textual component so that it jointly serves as speech guidance and reasoning representation, ECoM Reasoning improves reasoning accuracy while using a smaller token budget than the standard Chain-of-Modality (CoM) architecture, which generates intermediate text before speech. To train this capability, we further propose Progressive Compression, a curriculum-based strategy that gradually trains the model from full-form reasoning to compressed reasoning. Experiments on spoken mathematical question answering benchmarks show that ECoM Reasoning improves accuracy by 21% over standard CoM without explicit reasoning, and by 3% over CoM with full reasoning traces while using only 40% of the text tokens, demonstrating that it enhances SLM reasoning while remaining inference-efficient.

cs.CL↗

arXiv2026

SLAI T-Rex: Full-Parameter Post-training of the DeepSeek-V4 Family on Ascend SuperPOD

Full-parameter post-training of trillion-parameter-scale MoE models introduces substantial system-level challenges for large-scale distributed training, including severe memory pressure, non-overlapped communication overhead, and inefficient kernel execution. While most large-scale LLM training systems are built around GPU-based clusters, this report presents an end-to-end optimization practice on the Ascend NPU SuperPOD. Using the DeepSeek-V4 model family as the target workload, we develop a hierarchical optimization framework spanning model-level parallelism, computation-communication orchestration, and low-level kernel execution. The resulting system achieves 34.22% Model FLOPs Utilization (MFU) with a 2.93x improvement over the open-source baseline recipe while maintaining training stability. Building on this optimized infrastructure, we further establish a CPT and SFT workflow for complex Operations Research (OR) tasks. We refer to the integrated framework as SLAI T-Rex. Using DeepSeek-V4-Flash, we develop OR-oriented CPT and SFT data pipelines that combine collected domain resources with solver-verified synthetic optimization documents. The resulting dataset contains 10K high-quality SFT samples spanning four task categories and three problem representations. The specialized model achieves the highest average zero-shot Pass@1 score among the evaluated models, reaching 71.81% and outperforming GPT-5.4-Mini and the base DeepSeek-V4-Flash model by 3.98 and 11.27 percentage points, respectively. Overall, this work demonstrates a full-stack pathway from efficient trillion-parameter model post-training on Ascend infra to domain-specialized Flash models for solver-grounded mathematical modeling, advancing frontier-model systems for complex reasoning.

cs.CL↗

arXiv2026

High-accuracy ultrasonic positioning of calibration sources in the Jiangmen Underground Neutrino Observatory

Precise source positioning is essential for detector calibration in large liquid scintillator detectors such as JUNO, particularly in regions where purely mechanical control is insufficient. An ultrasonic positioning system has been developed to reconstruct the three-dimensional coordinates of a calibration source without interfering with photon collection or contaminating the liquid scintillator. The method combines a sound-speed modeling based on dedicated laboratory measurements and in-detector temperature profiles, waveform-based arrival-time reconstruction, and an in-situ calibration of the effective receiver geometry using central-axis deployments. With six active receivers, central-axis positioning yields a mean error of 1.23 cm relative to the known deployment reference. For off-axis operation in the Cable Loop System calibration plane, a detector-realistic simulation that includes timing resolution, sound-speed variation, and receiver-coordinate smearing predicts a positioning uncertainty of 2.40 cm. These results demonstrate that ultrasonic positioning can provide centimetre-level source accuracy for large liquid scintillator detectors and can support off-axis calibration in JUNO-like experiments.

physics.ins-det↗

arXiv2026

Radially correlated partially coherent beams with a deterministic vortex structure

Partially coherent beams have attracted considerable attention due to their intrinsic resilience against complex environmental perturbations. However, the intrinsic wavefront fluctuations make it fundamentally challenging to preserve well-defined orbital angular momentum during propagation. In this work, we propose and experimentally demonstrate a class of radially correlated, partially coherent beams that carry deterministic vortex structures, generated via optical conformal mapping from Cartesian to log-polar coordinates. The resulting beams exhibit a ring-shaped coherence distribution, characterized by low coherence in the radial direction and high coherence in the azimuthal direction. This unique feature of such a beam supports a well-defined deterministic vortex phase, thereby enabling the beam to preserve its ring-shaped coherence distribution during propagation through a focusing system. Our results provide new insights into the design of new partially coherent beams and may facilitate the development of applications in optical encoding, free-space information transmission, and ultrafast light-matter interactions.

physics.optics↗

arXiv2026

When Should Service Agents Reconsider? Difficulty-Routed Control in Customer-Service Operations

Autonomous customer-service agents are shifting from conversational interfaces toward operational execution roles: they retrieve firm records, apply service policies, and execute backend writes such as refunds, cancellations, exchanges, order modifications, and reservation changes. This shift creates a service-control problem: firms must keep routine service fast and low-friction while preventing operational errors on requests where customer instructions, policy constraints, firm records, and backend writes interact. We propose a difficulty-routed service-control architecture that asks when service agents should reconsider before acting. A lightweight router keeps routine sessions on a low-cost baseline path and routes operationally coupled sessions to an escalated workflow. The escalated path uses conflict-aware communication and write-triggered reconsideration to concentrate deliberation and safeguards before consequential backend writes, rather than applying additional control uniformly across all service sessions. We evaluate the architecture on human-verified retail and airline tasks from $\tau^{2}$-bench. In retail, the method improves reliability consistently on service requests with operational conflict. Routing evidence shows that stronger control is directed toward conflicted requests rather than broadly applied to routine ones. Dialogue and tool-use profiles suggest that gains do not come from indiscriminate interaction expansion or broader tool chains; instead, added turns and tool calls support evidence gathering, write separation, and pre-write reconsideration. Case-level evidence shows that the escalated workflow preserves fallback plans, binds retrieved records to the correct action, sequences writes, and decomposes multi-entity requests. Airline results extend the same service-control logic to reservation operations.

cs.AI↗

arXiv2026

Most Properties are Undecidable for Transitive Tense Logics

A logics' property is decidable in a class of logics if there exists an algorithm that decides whether a finitely axiomatizable logic in the class has the property. Many properties are undecidable for bimodal logics but decidable for linear tense logics, which leads to a general question on how the interactions of modalities affect the decidability of properties. In this paper, we study the decidability of properties for transitive tense logics and show that most properties are undecidable in the lattice NExt(K4t) of transitive tense logics, including Kripke completeness, the finite model property, and decidability. Our proof method adapts Chagrov's approach of constructing a reduction from an undecidable problem of Minsky machines to the decision problem for logics' properties, yielding a general scheme of proving the undecidability of these properties.

cs.LO↗