arXiv subjects

Zhu Zhang

Publications and source records attributed to Zhu Zhang.

At least 19 recordsLinked to original sources

arXiv2026

A*-Thought-V2: Efficient Latent Reasoning via Geometric Dynamics of LLM

Chain-of-Thought (CoT) improves the reasoning ability of Large Language Models (LLMs) but incurs substantial computation and context costs. Existing methods either lose intermediate information through hard pruning or lack a principled criterion for continuous compression. We present A*-Thought-V2, a geometric dynamics of LLM guided framework that models CoT as a hidden-state trajectory and replaces hard deletion with an explicit-implicit interleaved latent architecture. After projecting question, step, and solution representations into a 3D PCA space, it measures alignment between each local transition and global question-to-solution direction. Aligned steps remain explicit text, whereas deviating steps are compressed into continuous latent tokens. Directional angles capture both local semantics and reasoning dynamics: small angles indicate direct execution and answer formation, while large angles more frequently involve checking, correction, and branch exploration; their temporal variation reveals exploration, convergence, and refinement stages. To train this architecture, we introduce stepwise embedding forcing, which pools each redundant step into a single latent embedding, and label forcing, which supervises that latent token with a soft multi-modal vocabulary distribution instead of a hard one-hot label. Experiments on Qwen3.5-9B and Qwen3.6-27B across six in-domain and out-of-domain benchmarks show that A*-Thought-V2 improves average accuracy by up to 2.6% while reducing response length by up to half, increasing Accuracy per Computation Unit by 2.29$\times$, and reducing preprocessing and training time by 94.6% and up to 80.3%, respectively. Representation analyses suggest that latent states form a compact region distinct from textual states, while higher entropy at latent-token positions reflects broader soft targets that encourage richer step-level feature learning.

cs.CL↗

arXiv2026

Shock profiles for the cutoff Boltzmann equation of a binary gas mixture

We prove the existence of small-amplitude traveling shock profiles for the one-dimensional Boltzmann equation of a binary gas mixture with angular cutoff potentials in the full range $-3<\gamma\le 1$. The result extends the classical construction of Caflisch and Nicolaenko from hard potentials to the cutoff soft-potential regime. Indeed, the argument of proofs combines a Lyapunov--Schmidt reduction of the macroscopic component to a Burgers equation with an accelerated backward bi-characteristic method and a weighted $L^2$--$L^\infty$ iteration. Acceleration restores a uniformly positive collision frequency, compensating for the lack of a spectral gap for soft potentials, while the $L^2$--$L^\infty$ framework accommodates the absence of velocity smoothing induced by the cutoff, including a possible singularity along the grazing characteristic $v_1=s$. The shock profile tends to the Rankine--Hugoniot bi-Maxwellians at a mixed exponential rate as $|x|\to \infty$, with a sub-exponential remainder of order $|\varepsilon x|^{2/(3-\gamma)}$.

math.AP↗

arXiv2026

Wide-Field Opto-Iontronic iSCAT Mapping of Interfacial Charging and Electrical Connectivity

Spatial variations in electrical connectivity and interfacial ion accumulation can strongly influence electrochemical performance, yet these properties are difficult to visualize directly with conventional ensemble measurements. Here, we introduce potential-modulated opto-iontronic microscopy, an interferometric scattering microscopy (iSCAT) approach for wide-field imaging of electric-double-layer (EDL) dynamics at nanostructured electrodes. Sinusoidal potentials were applied to focused-ion-beam-fabricated indium tin oxide (ITO) nanoholes, and the optical response was extracted at the modulation frequency by Fourier demodulation. The optical modulation amplitude increased approximately linearly with modulation voltage above a low-voltage roll-off and decreased with increasing frequency, consistent with kinetically limited interfacial charging. We then mapped the potential-synchronized optical amplitude across patterned ITO electrodes. Electrically isolated blocks exhibited strongly suppressed modulation signals, whereas electrically connected and partially milled nanogrid structures showed pronounced responses. These results demonstrate label-free optical mapping of local charging dynamics and electrical connectivity at heterogeneous electrochemical interfaces using a commercially available iSCAT platform.

physics.chem-ph↗

arXiv2026

Beyond Teacher Likelihood: Group-Calibrated On-Policy Distillation for Long-Context Reasoning

On-policy distillation (OPD) trains a student on its own responses using dense token-level guidance from a stronger teacher. In long-context tasks, however, token-level teacher support can favor locally plausible responses that omit evidence distributed across the input or violate global task constraints. Task-specific verifiers, in contrast, evaluate task completion at the response level and may return graded rewards that reflect partial success. We diagnose this mismatch on fixed responses from two representative long-context evidence-aggregation tasks. Across longer input ranges, trajectory-level OPD scores become progressively less aligned with verifier rewards, indicating teacher-verifier disagreement. Motivated by this observation, we introduce Group-Calibrated On-Policy Distillation (GC-OPD). GC-OPD separately normalizes verifier rewards and trajectory-level OPD scores within each rollout group and uses their difference as a signed teacher-verifier disagreement residual. Relative-advantage-based credit assignment (RACA) distributes this trajectory-level residual across tokens according to their relative OPD advantages while preserving the original OPD signal. Across five long-context benchmarks, post-training with GC-OPD raises the five-benchmark averages of the official Qwen3-4B and Qwen3-8B checkpoints from 29.08 to 40.47 and from 35.12 to 44.65, respectively. Vanilla OPD reaches 39.31 and 43.56 under the same setup. Controlled ablations show that the signed residual is more effective than either an additional OPD-derived term or direct group-normalized verifier reward addition, while RACA further improves over uniform token allocation. Together, these results demonstrate that group-relative residual calibration can incorporate verifier outcomes without discarding dense token-level guidance. Code is available at https://github.com/SolereZhang/GC-OPD.

cs.LG↗

arXiv2026

MiniCPM-SALA: Hybridizing Sparse and Linear Attention for Efficient Long-Context Modeling

The evolution of large language models (LLMs) towards applications with ultra-long contexts faces challenges posed by the high computational and memory costs of the Transformer architecture. While existing sparse and linear attention mechanisms attempt to mitigate these issues, they typically involve a trade-off between memory efficiency and model performance. This paper introduces MiniCPM-SALA, a 9B-parameter hybrid architecture that integrates the high-fidelity long-context modeling of sparse attention (InfLLM-V2) with the global efficiency of linear attention (Lightning Attention). By employing a layer selection algorithm to integrate these mechanisms in a 1:3 ratio and utilizing a hybrid positional encoding (HyPE), the model maintains efficiency and performance for long-context tasks. Furthermore, we introduce a cost-effective continual training framework that transforms pre-trained Transformer-based models into hybrid models, which reduces training costs by approximately 75% compared to training from scratch. Extensive experiments show that MiniCPM-SALA maintains general capabilities comparable to full-attention models while offering improved efficiency. On a single NVIDIA A6000D GPU, the model achieves up to 3.5x the inference speed of the full-attention model at the sequence length of 256K tokens and supports context lengths of up to 1M tokens, a scale where traditional full-attention 8B models fail because of memory constraints.

cs.CL↗

arXiv2026

Hybrid Linear Attention Done Right: Efficient Distillation and Effective Architectures for Extremely Long Contexts

Hybrid Transformer architectures, which combine softmax attention blocks and recurrent neural networks (RNNs), have shown a desirable performance-throughput tradeoff for long-context modeling, but their adoption and studies are hindered by the prohibitive cost of large-scale pre-training from scratch. Some recent studies have shown that pre-trained softmax attention blocks can be converted into RNN blocks through parameter transfer and knowledge distillation. However, these transfer methods require substantial amounts of training data (more than 10B tokens), and the resulting hybrid models also exhibit poor long-context performance, which is the scenario where hybrid models enjoy significant inference speedups over Transformer-based models. In this paper, we present HALO (Hybrid Attention via Layer Optimization), a pipeline for distilling Transformer models into RNN-attention hybrid models. We then present HypeNet, a hybrid architecture with superior length generalization enabled by a novel position encoding scheme (named HyPE) and various architectural modifications. We convert the Qwen3 series into HypeNet using HALO, achieving performance comparable to the original Transformer models while enjoying superior long-context performance and efficiency. The conversion requires just 2.3B tokens, less than 0.01% of their pre-training data

cs.CL↗

arXiv2025

A Medical Multimodal Diagnostic Framework Integrating Vision-Language Models and Logic Tree Reasoning

With the rapid growth of large language models (LLMs) and vision-language models (VLMs) in medicine, simply integrating clinical text and medical imaging does not guarantee reliable reasoning. Existing multimodal models often produce hallucinations or inconsistent chains of thought, limiting clinical trust. We propose a diagnostic framework built upon LLaVA that combines vision-language alignment with logic-regularized reasoning. The system includes an input encoder for text and images, a projection module for cross-modal alignment, a reasoning controller that decomposes diagnostic tasks into steps, and a logic tree generator that assembles stepwise premises into verifiable conclusions. Evaluations on MedXpertQA and other benchmarks show that our method improves diagnostic accuracy and yields more interpretable reasoning traces on multimodal tasks, while remaining competitive on text-only settings. These results suggest a promising step toward trustworthy multimodal medical AI.

cs.AI↗

arXiv2025

LLM$\times$MapReduce-V3: Enabling Interactive In-Depth Survey Generation through a MCP-Driven Hierarchically Modular Agent System

We introduce LLM x MapReduce-V3, a hierarchically modular agent system designed for long-form survey generation. Building on the prior work, LLM x MapReduce-V2, this version incorporates a multi-agent architecture where individual functional components, such as skeleton initialization, digest construction, and skeleton refinement, are implemented as independent model-context-protocol (MCP) servers. These atomic servers can be aggregated into higher-level servers, creating a hierarchically structured system. A high-level planner agent dynamically orchestrates the workflow by selecting appropriate modules based on their MCP tool descriptions and the execution history. This modular decomposition facilitates human-in-the-loop intervention, affording users greater control and customization over the research process. Through a multi-turn interaction, the system precisely captures the intended research perspectives to generate a comprehensive skeleton, which is then developed into an in-depth survey. Human evaluations demonstrate that our system surpasses representative baselines in both content depth and length, highlighting the strength of MCP-based modular planning.

cs.CL↗

arXiv2025

MedLA: A Logic-Driven Multi-Agent Framework for Complex Medical Reasoning with Large Language Models

Answering complex medical questions requires not only domain expertise and patient-specific information, but also structured and multi-perspective reasoning. Existing multi-agent approaches often rely on fixed roles or shallow interaction prompts, limiting their ability to detect and resolve fine-grained logical inconsistencies. To address this, we propose \textsc{MedLA}, a logic-driven multi-agent framework built on large language models. Each agent organizes its reasoning process into an explicit logical tree based on syllogistic triads (major premise, minor premise, and conclusion), enabling transparent inference and premise-level alignment. Agents engage in a multi-round, graph-guided discussion to compare and iteratively refine their logic trees, achieving consensus through error correction and contradiction resolution. We demonstrate that \textsc{MedLA} consistently outperforms both static role-based systems and single-agent baselines on challenging benchmarks such as MedDDx and standard medical QA tasks. Furthermore, \textsc{MedLA} scales effectively across both open-source and commercial LLM backbones, achieving state-of-the-art performance and offering a generalizable paradigm for trustworthy medical reasoning.

cs.AI↗

arXiv2025

Unstable mode around the 3D boundary layer flow

We study the stability properties of boundary layer-type shear flows for the three-dimensional Navier-Stokes equations in the limit of small viscosity $0<\nu\ll 1$. When the streamwise and spanwise velocity profiles are linearly independent near the boundary, we construct an unstable mode that exhibits rapid growth at the rate of $e^{t/\sqrt{\nu}}$. Our results reveal an analytic instability in the three-dimensional Navier-Stokes equations around generic boundary layer profiles. This instability arises from the interplay between spanwise flow and three-dimensional perturbations, and does not occur in purely two-dimensional flows.

math.AP↗

arXiv2025

Optical Voltammetry of redox processes inside a nanohole with opto-iontronic microscopy

Cyclic Voltammetry (CV) is the most commonly used method in electrochemistry to characterize electrochemical reactions, usually involving macroscopic electrodes. Here we demonstrate an optical CV technique called Opto-iontronic Microscopy, which is capable of monitoring electrochemical processes at the nanoscale. By integrating optical microscopy with nanohole electrodes, we enhance sensitivity in detecting redox reactions within volumes as small as an attoliter ($(100 \text{~nm})^{3}$). This technique uses total internal reflection illumination, Electric-double-layer modulation, cyclic voltammetry, and lock-in detection, to probe ion dynamics inside nanoholes. We applied this method to study EDL (dis)charging coupled to ferrocenedimethanol (Fc(MeOH)$_2$) redox reactions. Experimental results were validated against a theoretical Poisson-Nernst-Planck-Butler-Volmer model, providing insights into ion concentration changes of reaction species that contribute to the optical contrast. This work opens up opportunities for high-sensitivity, label-free analysis of electrochemical reactions in nanoconfined environments, with potential applications in pure nanocrystal growth and monitoring.

cond-mat.soft↗

arXiv2025

Monocle: Hybrid Local-Global In-Context Evaluation for Long-Text Generation with Uncertainty-Based Active Learning

Assessing the quality of long-form, model-generated text is challenging, even with advanced LLM-as-a-Judge methods, due to performance degradation as input length increases. To address this issue, we propose a divide-and-conquer approach, which breaks down the comprehensive evaluation task into a series of localized scoring tasks, followed by a final global assessment. This strategy allows for more granular and manageable evaluations, ensuring that each segment of the text is assessed in isolation for both coherence and quality, while also accounting for the overall structure and consistency of the entire piece. Moreover, we introduce a hybrid in-context learning approach that leverages human annotations to enhance the performance of both local and global evaluations. By incorporating human-generated feedback directly into the evaluation process, this method allows the model to better align with human judgment. Finally, we develop an uncertainty-based active learning algorithm that efficiently selects data samples for human annotation, thereby reducing annotation costs in practical scenarios. Experimental results show that the proposed evaluation framework outperforms several representative baselines, highlighting the effectiveness of our approach.

cs.CL↗

arXiv2025

LLM$\times$MapReduce-V2: Entropy-Driven Convolutional Test-Time Scaling for Generating Long-Form Articles from Extremely Long Resources

Long-form generation is crucial for a wide range of practical applications, typically categorized into short-to-long and long-to-long generation. While short-to-long generations have received considerable attention, generating long texts from extremely long resources remains relatively underexplored. The primary challenge in long-to-long generation lies in effectively integrating and analyzing relevant information from extensive inputs, which remains difficult for current large language models (LLMs). In this paper, we propose LLM$\times$MapReduce-V2, a novel test-time scaling strategy designed to enhance the ability of LLMs to process extremely long inputs. Drawing inspiration from convolutional neural networks, which iteratively integrate local features into higher-level global representations, LLM$\times$MapReduce-V2 utilizes stacked convolutional scaling layers to progressively expand the understanding of input materials. Both quantitative and qualitative experimental results demonstrate that our approach substantially enhances the ability of LLMs to process long inputs and generate coherent, informative long-form articles, outperforming several representative baselines. Both LLM$\times$MapReduce-V2 and SurveyEval are publicly available at https://github.com/thunlp/LLMxMapReduce .

cs.CL↗

arXiv2025

Structural stability of boundary layers in the entire subsonic regime

Despite the physical importance, there are limited mathematical theories for the compressible Navier-Stokes equations with strong boundary layers. This is mainly due to the absence of a stream function structure, unlike the extensively studied incompressible fluid dynamics in two dimensions. This paper aims to establish the structural stability of boundary layer profiles in the form of shear flow for the two-dimensional steady compressible Navier-Stokes equations. Our estimates are uniform across the entire subsonic regime, where the Mach number $m\in (0,1)$. As a byproduct, we provide the first result concerning the low Mach number limit in the presence of Prandtl boundary layers. The proof relies on the quasi-compressible-Stokes iteration introduced in [38], along with a subtle analysis of the interplay between density and velocity variables in different frequency regimes, and the identification of cancellations in higher-order estimates.

math.AP↗

arXiv2024

Laser-patterned Thin-film Electrodes: Imaging Ion Accumulation and Trapped Nanoparticles

This study introduces a straightforward electrode design featuring sharp edges with a curvature of a few hundred nanometers in radius, with which both ion accumulation and nanoparticle deposition can be observed under an alternating electrical potential. The electrodes, termed 'shark-teeth electrodes', are fabricated using a laser ablation technique optimized for facile nanostructure creation. This method involves successive, overlapping ablated discs in a thin film of gold, producing sharp tips that generate strong electric fields. When electrically polarized in an electrolyte solution, these sharp tips form a screening layer, facilitating the observation of ion and nanoparticle behavior. A total-internal reflection microscope is employed to monitor ion accumulation on these electrodes, demonstrating their capability in iontronic microscopy. Additionally, the same electrodes are used to track nanoparticle trapping under high-frequency alternating potentials. This dual functionality allows for the investigation of electrochemical and physical interactions between ions and colloidal nanoparticles, contributing valuable insights to the field of soft matter.

cond-mat.soft↗

arXiv2024

Few-Shot Learning for Mental Disorder Detection: A Continuous Multi-Prompt Engineering Approach with Medical Knowledge Injection

This study harnesses state-of-the-art AI technology for detecting mental disorders through user-generated textual content. Existing studies typically rely on fully supervised machine learning, which presents challenges such as the labor-intensive manual process of annotating extensive training data for each research problem and the need to design specialized deep learning architectures for each task. We propose a novel method to address these challenges by leveraging large language models and continuous multi-prompt engineering, which offers two key advantages: (1) developing personalized prompts that capture each user's unique characteristics and (2) integrating structured medical knowledge into prompts to provide context for disease detection and facilitate predictive modeling. We evaluate our method using three widely prevalent mental disorders as research cases. Our method significantly outperforms existing methods, including feature engineering, architecture engineering, and discrete prompt engineering. Meanwhile, our approach demonstrates success in few-shot learning, i.e., requiring only a minimal number of training examples. Moreover, our method can be generalized to other rare mental disorder detection tasks with few positive labels. In addition to its technical contributions, our method has the potential to enhance the well-being of individuals with mental disorders and offer a cost-effective, accessible alternative for stakeholders beyond traditional mental disorder screening methods.

cs.CL↗

arXiv2023

Weakly-Supervised Video Moment Retrieval via Regularized Two-Branch Proposal Networks with Erasing Mechanism

Video moment retrieval is to identify the target moment according to the given sentence in an untrimmed video. Due to temporal boundary annotations of the video are extremely time-consuming to acquire, modeling in the weakly-supervised setting is increasingly focused, where we only have access to the video-sentence pairs during training. Most existing weakly-supervised methods adopt a MIL-based framework to develop inter-sample confrontment, but neglect the intra-sample confrontment between moments with similar semantics. Therefore, these methods fail to distinguish the correct moment from plausible negative moments. Further, the previous attention models in cross-modal interaction tend to focus on a few dominant words exorbitantly, ignoring the comprehensive video-sentence correspondence. In this paper, we propose a novel Regularized Two-Branch Proposal Network with Erasing Mechanism to consider the inter-sample and intra-sample confrontments simultaneously. Concretely, we first devise a language-aware visual filter to generate both enhanced and suppressed video streams. Then, we design the sharable two-branch proposal module to generate positive and plausible negative proposals from the enhanced and suppressed branch respectively, contributing to sufficient confrontment. Besides, we introduce an attention-guided dynamic erasing mechanism in enhanced branch to discover the complementary video-sentence relation. Moreover, we apply two types of proposal regularization to stabilize the training process and improve model performance. The extensive experiments on ActivityCaption, Charades-STA and DiDeMo datasets show the effectiveness of our method.

cs.MM↗

arXiv2023

Long-time instability of planar Poiseuille-type flow in compressible fluid

It is well-known that at the high Reynolds number, the linearized Navier-Stokes equations around the inviscid stable shear profile admit growing mode solutions due to the destabilizing effect of the viscosity. This phenomenon, called Tollmien-Schlichting instability, has been rigorously justified by Grenier-Guo-Nguyen [Adv. Math. 292 (2016); Duke J. Math. 165 (2016)] for Poiseuille flows and boundary layers in the incompressible fluid. To reveal this intrinsic instability mechanism in the compressible setting, in this paper, we study the long-time instability of the Poiseuille flow in a channel. Note that this instability arises in a low-frequency regime instead of a high-frequency regime for the Prandtl boundary layer. The proof is based on the quasi-compressible-Stokes iteration introduced by Yang-Zhang in [50] and subtle analysis of the dispersion relation for the instability. Note that we do not require symmetric conditions on the background shear flow or perturbations.

math.AP↗