arXiv subjects

Zhenyu Zhang

Publications and source records attributed to Zhenyu Zhang.

At least 109 records · Page 6Linked to original sources

arXiv2025

COOPER: A Unified Model for Cooperative Perception and Reasoning in Spatial Intelligence

Visual Spatial Reasoning is crucial for enabling Multimodal Large Language Models (MLLMs) to understand object properties and spatial relationships, yet current models still struggle with 3D-aware reasoning. Existing approaches typically enhance either perception, by augmenting RGB inputs with auxiliary modalities such as depth and segmentation, or reasoning, by training on spatial VQA datasets and applying reinforcement learning, and thus treat these two aspects in isolation. In this work, we investigate whether a unified MLLM can develop an intrinsic ability to enhance spatial perception and, through adaptive interleaved reasoning, achieve stronger spatial intelligence. We propose \textbf{COOPER}, a unified MLLM that leverages depth and segmentation as auxiliary modalities and is trained in two stages to acquire auxiliary modality generation and adaptive, interleaved reasoning capabilities. COOPER achieves an average \textbf{6.91\%} improvement in spatial reasoning while maintaining general performance. Moreover, even a variant trained only for auxiliary modality generation attains a \textbf{7.92\%} gain on distance and size estimation, suggesting that learning to generate auxiliary modalities helps internalize spatial knowledge and strengthen spatial understanding.

cs.CV↗

arXiv2025

V-ITI: Mitigating Hallucinations in Multimodal Large Language Models via Visual Inference-Time Intervention

Multimodal Large Language Models (MLLMs) excel in numerous vision-language tasks yet suffer from hallucinations, producing content inconsistent with input visuals, that undermine reliability in precision-sensitive domains. This issue stems from a fundamental problem of visual neglect, where models fail to adequately prioritize input images. Existing methods typically alleviate hallucinations by intervening in the attention score or output logits, focusing on "how to intervene" but overlooking the prerequisite "when to intervene", which leads to the "over-intervention" problem and subsequently introduces new hallucinations and unnecessary computational overhead. To address this gap, we first investigate the mechanism of visual neglect and reveal it can be accurately detected via head-level activation patterns in MLLMs. We thus propose V-ITI, a lightweight visual inference-time intervention framework integrating a Visual Neglect Detector that identifies visual neglect via head-level discriminative probes and a Visual Recall Intervenor that modulates activations with prestored visual activation information only when the visual neglect is detected. Extensive experiments across eight benchmarks and different MLLM families demonstrate that V-ITI consistently mitigates vision-related hallucinations while preserving general task performance.

cs.CV↗

arXiv2025

Warm Chat: Diffuse Emotion-aware Interactive Talking Head Avatar with Tree-Structured Guidance

Generative models have advanced rapidly, enabling impressive talking head generation that brings AI to life. However, most existing methods focus solely on one-way portrait animation. Even the few that support bidirectional conversational interactions lack precise emotion-adaptive capabilities, significantly limiting their practical applicability. In this paper, we propose Warm Chat, a novel emotion-aware talking head generation framework for dyadic interactions. Leveraging the dialogue generation capability of large language models (LLMs, e.g., GPT-4), our method produces temporally consistent virtual avatars with rich emotional variations that seamlessly transition between speaking and listening states. Specifically, we design a Transformer-based head mask generator that learns temporally consistent motion features in a latent mask space, capable of generating arbitrary-length, temporally consistent mask sequences to constrain head motions. Furthermore, we introduce an interactive talking tree structure to represent dialogue state transitions, where each tree node contains information such as child/parent/sibling nodes and the current character's emotional state. By performing reverse-level traversal, we extract rich historical emotional cues from the current node to guide expression synthesis. Extensive experiments demonstrate the superior performance and effectiveness of our method.

eess.AS↗

arXiv2025

A Comprehensive Regime Diagram of Dynamical Modes of Triple Flickering Buoyant Diffusion Flames: Experimental and Model Investigations

The triple-flame system serves as the fundamental unit for understanding multi-flame interactions, revealing critical coupling mechanisms that scale to complex burner arrays. In this study, we investigated triple flame oscillators, consisting of three flickering laminar buoyant diffusion flames arranged in an isosceles triangular configuration, to construct a comparative regime diagram of dynamical modes. To overcome the limited experimental observability caused by the discretization of geometric parameters, we enabled continuous motion of the vertex flame at a controlled speed V, while independently varying the base length L and the fuel flow rate Q. We conducted a systematic investigation of the triple flame coupling behaviors by varying the triangle size, fuel flow rate, and vertex flame movement velocity. Based on the experimental observations, a comprehensive regime diagram was established to classify the dynamical modes of triple flickering buoyant diffusion flames. Notably, three previously unreported dynamical modes were identified for the first time. To interpret these complex flame interactions, a Stuart-Landau oscillator model with time-delay coupling was employed, which successfully reproduces the experimentally observed dynamical modes. Experimentally observed dynamical modes reveal a bifurcation diagram for the coupled triple Stuart-Landau system, elucidating the transitions between different synchronization modes.

physics.flu-dyn↗

arXiv2025

Dual-Kernel Graph Community Contrastive Learning

Graph Contrastive Learning (GCL) has emerged as a powerful paradigm for training Graph Neural Networks (GNNs) in the absence of task-specific labels. However, its scalability on large-scale graphs is hindered by the intensive message passing mechanism of GNN and the quadratic computational complexity of contrastive loss over positive and negative node pairs. To address these issues, we propose an efficient GCL framework that transforms the input graph into a compact network of interconnected node sets while preserving structural information across communities. We firstly introduce a kernelized graph community contrastive loss with linear complexity, enabling effective information transfer among node sets to capture hierarchical structural information of the graph. We then incorporate a knowledge distillation technique into the decoupled GNN architecture to accelerate inference while maintaining strong generalization performance. Extensive experiments on sixteen real-world datasets of varying scales demonstrate that our method outperforms state-of-the-art GCL baselines in both effectiveness and scalability.

cs.LG↗

arXiv2025

The Path Not Taken: RLVR Provably Learns Off the Principals

Reinforcement Learning with Verifiable Rewards (RLVR) reliably improves the reasoning performance of large language models, yet it appears to modify only a small fraction of parameters. We revisit this paradox and show that sparsity is a surface artifact of a model-conditioned optimization bias: for a fixed pretrained model, updates consistently localize to preferred parameter regions, highly consistent across runs and largely invariant to datasets and RL recipes. We mechanistically explain these dynamics with a Three-Gate Theory: Gate I (KL Anchor) imposes a KL-constrained update; Gate II (Model Geometry) steers the step off principal directions into low-curvature, spectrum-preserving subspaces; and Gate III (Precision) hides micro-updates in non-preferred regions, making the off-principal bias appear as sparsity. We then validate this theory and, for the first time, provide a parameter-level characterization of RLVR's learning dynamics: RLVR learns off principal directions in weight space, achieving gains via minimal spectral drift, reduced principal-subspace rotation, and off-principal update alignment. In contrast, SFT targets principal weights, distorts the spectrum, and even lags RLVR. Together, these results provide the first parameter-space account of RLVR's training dynamics, revealing clear regularities in how parameters evolve. Crucially, we show that RL operates in a distinct optimization regime from SFT, so directly adapting SFT-era parameter-efficient fine-tuning (PEFT) methods can be flawed, as evidenced by our case studies on advanced sparse fine-tuning and LoRA variants. We hope this work charts a path toward a white-box understanding of RLVR and the design of geometry-aware, RLVR-native learning algorithms, rather than repurposed SFT-era heuristics.

cs.LG↗

arXiv2025

Bridging Tool Dependencies and Domain Knowledge: A Graph-Based Framework for In-Context Planning

We present a framework for uncovering and exploiting dependencies among tools and documents to enhance exemplar artifact generation. Our method begins by constructing a tool knowledge graph from tool schemas,including descriptions, arguments, and output payloads, using a DeepResearch-inspired analysis. In parallel, we derive a complementary knowledge graph from internal documents and SOPs, which is then fused with the tool graph. To generate exemplar plans, we adopt a deep-sparse integration strategy that aligns structural tool dependencies with procedural knowledge. Experiments demonstrate that this unified framework effectively models tool interactions and improves plan generation, underscoring the benefits of linking tool graphs with domain knowledge graphs for tool-augmented reasoning and planning.

cs.AI↗

arXiv2025

StorageXTuner: An LLM Agent-Driven Automatic Tuning Framework for Heterogeneous Storage Systems

Automatically configuring storage systems is hard: parameter spaces are large and conditions vary across workloads, deployments, and versions. Heuristic and ML tuners are often system specific, require manual glue, and degrade under changes. Recent LLM-based approaches help but usually treat tuning as a single-shot, system-specific task, which limits cross-system reuse, constrains exploration, and weakens validation. We present StorageXTuner, an LLM agent-driven auto-tuning framework for heterogeneous storage engines. StorageXTuner separates concerns across four agents - Executor (sandboxed benchmarking), Extractor (performance digest), Searcher (insight-guided configuration exploration), and Reflector (insight generation and management). The design couples an insight-driven tree search with layered memory that promotes empirically validated insights and employs lightweight checkers to guard against unsafe actions. We implement a prototype and evaluate it on RocksDB, LevelDB, CacheLib, and MySQL InnoDB with YCSB, MixGraph, and TPC-H/C. Relative to out-of-the-box settings and to ELMo-Tune, StorageXTuner reaches up to 575% and 111% higher throughput, reduces p99 latency by as much as 88% and 56%, and converges with fewer trials.

cs.DB↗

arXiv2025

ReCAP: Recursive Context-Aware Reasoning and Planning for Large Language Model Agents

Long-horizon tasks requiring multi-step reasoning and dynamic re-planning remain challenging for large language models (LLMs). Sequential prompting methods are prone to context drift, loss of goal information, and recurrent failure cycles, while hierarchical prompting methods often weaken cross-level continuity or incur substantial runtime overhead. We introduce ReCAP (Recursive Context-Aware Reasoning and Planning), a hierarchical framework with shared context for reasoning and planning in LLMs. ReCAP combines three key mechanisms: (i) plan-ahead decomposition, in which the model generates a full subtask list, executes the first item, and refines the remainder; (ii) structured re-injection of parent plans, maintaining consistent multi-level context during recursive return; and (iii) memory-efficient execution, bounding the active prompt so costs scale linearly with task depth. Together these mechanisms align high-level goals with low-level actions, reduce redundant prompting, and preserve coherent context updates across recursion. Experiments demonstrate that ReCAP substantially improves subgoal alignment and success rates on various long-horizon reasoning benchmarks, achieving a 32% gain on synchronous Robotouille and a 29% improvement on asynchronous Robotouille under the strict pass@1 protocol.

cs.AI↗

arXiv2025

SEAL: Steerable Reasoning Calibration of Large Language Models for Free

Large Language Models (LLMs), such as OpenAI's o1-series have demonstrated compelling capabilities for complex reasoning tasks via the extended chain-of-thought (CoT) reasoning mechanism. However, recent studies reveal substantial redundancy in the CoT reasoning traces, which not only increases inference latency but also negatively impacts model performance by diverting attention to unnecessary reasoning paths. To address this issue, we investigate the internal reasoning structures of LLMs and categorize them into three primary thought types: execution, reflection, and transition thoughts. Moreover, our analysis reveals that excessive reflection and transition thoughts are strongly correlated with failure cases and these thought categories exhibit clear separation in the latent space. Based on these, we introduce SEAL (Steerable reasoning calibration), a training-free approach that seamlessly calibrates the CoT process, improving accuracy while demonstrating significant efficiency gains. SEAL consists of an offline stage for extracting the reasoning steering vector in the latent space, followed by an on-the-fly calibration of the reasoning trace through representation intervention using the steering vector. Notably, the steering vector exhibits strong transferability across various tasks. Extensive experiments across multiple models (DeepSeek-R1-Distill and QwQ-32B-Preview) and benchmarks (Math500, GSM8K, LiveCodeBench) validate the effectiveness of SEAL, up to a 11% improvement in accuracy while reducing reasoning tokens by 11.8% to 50.4%. Our code is publicly available at https://github.com/VITA-Group/SEAL.

cs.CL↗

arXiv2025

Towards Robust and Realible Multimodal Misinformation Recognition with Incomplete Modality

Multimodal Misinformation Recognition has become an urgent task with the emergence of huge multimodal fake content on social media platforms. Previous studies mainly focus on complex feature extraction and fusion to learn discriminative information from multimodal content. However, in real-world applications, multimedia news may naturally lose some information during dissemination, resulting in modality incompleteness, which is detrimental to the generalization and robustness of existing models. To this end, we propose a novel generic and robust multimodal fusion strategy, termed Multi-expert Modality-incomplete Learning Network (MMLNet), which is simple yet effective. It consists of three key steps: (1) Multi-Expert Collaborative Reasoning to compensate for missing modalities by dynamically leveraging complementary information through multiple experts. (2) Incomplete Modality Adapters compensates for the missing information by leveraging the new feature distribution. (3) Modality Missing Learning leveraging an label-aware adaptive weighting strategy to learn a robust representation with contrastive learning. We evaluate MMLNet on three real-world benchmarks across two languages, demonstrating superior performance compared to state-of-the-art methods while maintaining relative simplicity. By ensuring the accuracy of misinformation recognition in incomplete modality scenarios caused by information propagation, MMLNet effectively curbs the spread of malicious misinformation. Code is publicly available at https://github.com/zhyhome/MMLNet.

cs.MM↗

arXiv2025

TCDiff++: An End-to-end Trajectory-Controllable Diffusion Model for Harmonious Music-Driven Group Choreography

Music-driven dance generation has garnered significant attention due to its wide range of industrial applications, particularly in the creation of group choreography. During the group dance generation process, however, most existing methods still face three primary issues: multi-dancer collisions, single-dancer foot sliding and abrupt swapping in the generation of long group dance. In this paper, we propose TCDiff++, a music-driven end-to-end framework designed to generate harmonious group dance. Specifically, to mitigate multi-dancer collisions, we utilize a dancer positioning embedding to encode temporal and identity information. Additionally, we incorporate a distance-consistency loss to ensure that inter-dancer distances remain within plausible ranges. To address the issue of single-dancer foot sliding, we introduce a swap mode embedding to indicate dancer swapping patterns and design a Footwork Adaptor to refine raw motion, thereby minimizing foot sliding. For long group dance generation, we present a long group diffusion sampling strategy that reduces abrupt position shifts by injecting positional information into the noisy input. Furthermore, we integrate a Sequence Decoder layer to enhance the model's ability to selectively process long sequences. Extensive experiments demonstrate that our TCDiff++ achieves state-of-the-art performance, particularly in long-duration scenarios, ensuring high-quality and coherent group dance generation.

cs.SD↗

arXiv2025

Autocorrelation signatures in time-resolved black hole flare images: secondary peaks and convergence structure

The strong gravitational field of a black hole bends light, forming multi-level images, yet extracting precise spacetime information from them remains challenging. In this study, we investigate how gravitational lensing leaves unique and detectable signatures in black hole movies using autocorrelation analysis. By examining the two-dimensional autocorrelation of a movie depicting a hotspot orbiting a Kerr black hole, as viewed by a near-axis observer, we identify a persistent secondary peak structure induced by gravitational lensing. Notably, these secondary peaks converge toward an approximately fixed point in the time-angle lag domain, largely independent of the orbital radius of the hotspot. This key property suggests that combining future flare observations with precise autocorrelation analysis could effectively disentangle lensing effects from orbital dynamics, enabling direct measurement of black hole parameters.

astro-ph.HE↗

arXiv2025

Optimally Tensile Strained La3Ni2O7 Films as Candidate High-Temperature Superconductors on Designer Ba1-xSrxO (001) and SrO-SrTiO3 Substrates

Recent experiments have observed superconductivity up to 48 K in La3Ni2O7-derived films under compressive strain imposed by the SrLaAlO4 substrate, while such films on the SrTiO3 substrate with tensile strain have failed to reach the superconducting state. Here we propose to broadly expand the choices of materials platforms to achieve high-Tc superconducting La3Ni2O7 films by proposing designer substrates of Ba1-xSrxO (x = 0 - 1) that allow to continuously tune the strain in the films from being tensile to compressive. Our systematic study of the structural and electronic reconstructions of the strained La3Ni2O7 bilayer film leads to the central finding that at the optimal tensile strain of ~2% (x ~0.25), the spectral weight of the Ni dz2 orbital is peaked right at the Fermi level, and its hybridization with the Ni dx2-y2 orbital is substantially enhanced. Consequently, the expected Tc should be unprecedentedly high, at least substantially higher than those achieved in the compressive regime. Furthermore, our detailed thickness-dependent energetic analyses show that such films can be stably grown for thicknesses equal to or beyond the bilayer regime, and predict that the SrO-terminated SrTiO3 should also be able to stabilize the films with optimal tensile strain and higher Tc's.

cond-mat.supr-con↗

arXiv2025

Computation of cohomology of finite Lie conformal algebras

By using the energy operator introduced by B. Bakalov, A. De Sole, and V.\,G. Kac, we propose an algorithm for computing the cohomology of finitely generated free Lie conformal algebras with a Virasoro element. This computational method enables us to determine cohomologies with coefficients in their conformal modules, including the trivial module, irreducible finitely free modules, and adjoint modules. As concrete applications, we compute by \textit{Mathematica} both basic and reduced cohomologies for the $\mathcal{W}(b)$, Schrödinger-Virasoro and extended Schrödinger-Virasoro Lie conformal algebras.

math.RA↗

arXiv2025

Describe, Don't Dictate: Semantic Image Editing with Natural Language Intent

Despite the progress in text-to-image generation, semantic image editing remains a challenge. Inversion-based algorithms unavoidably introduce reconstruction errors, while instruction-based models mainly suffer from limited dataset quality and scale. To address these problems, we propose a descriptive-prompt-based editing framework, named DescriptiveEdit. The core idea is to re-frame `instruction-based image editing' as `reference-image-based text-to-image generation', which preserves the generative power of well-trained Text-to-Image models without architectural modifications or inversion. Specifically, taking the reference image and a prompt as input, we introduce a Cross-Attentive UNet, which newly adds attention bridges to inject reference image features into the prompt-to-edit-image generation process. Owing to its text-to-image nature, DescriptiveEdit overcomes limitations in instruction dataset quality, integrates seamlessly with ControlNet, IP-Adapter, and other extensions, and is more scalable. Experiments on the Emu Edit benchmark show it improves editing accuracy and consistency.

cs.CV↗

arXiv2025

MARS-FTCP: Robust Fault-Tolerant Control and Agile Trajectory Planning for Modular Aerial Robot Systems

Modular Aerial Robot Systems (MARS) consist of multiple drone units that can self-reconfigure to adapt to various mission requirements and fault conditions. However, existing fault-tolerant control methods exhibit significant oscillations during docking and separation, impacting system stability. To address this issue, we propose a novel fault-tolerant control reallocation method that adapts to an arbitrary number of modular robots and their assembly formations. The algorithm redistributes the expected collective force and torque required for MARS to individual units according to their moment arm relative to the center of MARS mass. Furthermore, we propose an agile trajectory planning method for MARS of arbitrary configurations, which is collision-avoiding and dynamically feasible. Our work represents the first comprehensive approach to enable fault-tolerant and collision avoidance flight for MARS. We validate our method through extensive simulations, demonstrating improved fault tolerance, enhanced trajectory tracking accuracy, and greater robustness in cluttered environments. The videos and source code of this work are available at https://github.com/RuiHuangNUS/MARS-FTCP/

cs.RO↗

arXiv2025

Machine Learning Unveils the Power Law of Finite-Volume Energy Shifts

Finite-volume extrapolation is an important step for extracting physical observables from lattice calculations. However, it is a significant challenge for the system with long-range interactions. We employ symbolic regression to regress finite-volume extrapolation formula for both short-range and long-range interactions. The regressed formula still holds the exponential form with a factor $L^n$ in front of it. The power decreases with the decreasing range of the force. When the range of the force becomes sufficiently small, the power converges to $-1$, recovering the short-range formula as expected. Our work represents a significant advancement in leveraging machine learning to probe uncharted territories within particle physics.

hep-ph↗