arXiv subjects

Yizhou Zhou

Publications and source records attributed to Yizhou Zhou.

At least 19 recordsLinked to original sources

arXiv2026

HMARS: A Hierarchical Multi-Agent Memory System for Long-Context Reasoning

Long-context reasoning requires models to access, retrieve, and integrate evidence scattered across documents, dialogues, and accumulated interaction histories. Standard retrieval-augmented generation reduces this problem to top-$K$ chunk retrieval, but such passive access can discard relevant evidence before reasoning begins, especially when relevance depends on broader context. We propose HMARS, a hierarchical multi-agent memory system that treats long contexts as managed memory rather than a flat retrieval corpus. Sub-agents maintain grounded access to bounded memory regions, mid-agents manage regional context and provide query-specific coordination, and a frontier model performs final reasoning over retrieved evidence pages. To evaluate this view, we construct two diagnostic benchmarks targeting evidence breadth and context-dependent relevance. Across long-document and multi-turn memory tasks, HMARS achieves the best overall performance against retrieval, reranking, full-context, graph-based, and agentic long-context baselines. Evidence coverage analysis further shows that its gains come from retrieving the required supporting evidence more completely, rather than merely changing the final answer prompt.

cs.IR↗

arXiv2026

ECNUClaw: A Learner-Profiled Intelligent Study Companion Framework for K-12 Personalized Education

We introduce ECNUClaw, an open-source framework for building learner-profiled intelligent study companions in K-12 education. The system constructs and maintains a five-dimension learner profile -- covering cognitive, behavioral, emotional, metacognitive, and contextual dimensions -- by extracting signals from student-companion dialogues at each turn. Profile updates feed directly into an adaptive strategy engine that adjusts the companion's guidance intensity, encouragement frequency, and Bloom's taxonomy scaffolding in real time. The framework design draws on three theoretical strands from the Chinese educational technology literature: Zhang's Digital Portrait Three-Layer Framework for learner assessment, the Education Brain model for educational system architecture, and the Human-AI Collaborative IQ concept for companion design philosophy. ECNUClaw is implemented in Python and supports seven Chinese LLM providers through a unified OpenAI-compatible adapter layer. We describe the system architecture, the profiling and adaptation mechanisms, and discuss limitations and next steps. The source code is available at https://github.com/bushushu2333/ECNUClaw.

cs.HC↗

arXiv2026

Computation of entanglement for quantum states by a Consensus-Based Optimization method

The computation of quantum entanglement can be formulated as a high-dimensional nonconvex optimization problem with orthogonality constraints. In this work, we propose structure-preserving consensus-based optimization (CBO) methods for entanglement computation, with one approach based on a Hermitian formulation and the other evolving directly on the unitary manifold. To handle the variable dimension of the feasible set, we introduce a cross-dimensional interaction mechanism allowing exchange of information between particles of different sizes. Numerical experiments demonstrate that the proposed methods achieve accurate approximations.

quant-ph↗

arXiv2026

Error Estimates for Hyperbolic Scaling Limits of Linear Kinetic Models on Networks

This paper studies linear discrete kinetic models on networks and their asymptotic behavior in the small Knudsen number limit. For coupling conditions at an n-edge junction under a symmetric formulation, we introduce a change of variables that reformulates the system into n independent initial-boundary value problems. The asymptotic expansions are then constructed and rigorously justified by deriving an error estimate based on the energy method.

math.NA↗

arXiv2026

A spectral approach to interface layers on networks for the linearized BGK equation and its acoustic limit

We consider in this paper a velocity discretized version of the full linear kinetic BGK model and the corresponding limit for small Knudsen number, the linearised Euler or acoustic system. Considering these equations on networks, coupling conditions for the macroscopic equations are derived from the kinetic conditions via an asymptotic analysis near the nodes of the network. Here, a degeneracy in the limit equations requires not only the investigation of kinetic layers, but also the discussion of viscous layers. Using the kinetic coupling conditions at the junction and coupling kinetic and viscous layers to the outer problems on the edges one obtains a coupled kinetic half-space problem at each node. A spectral method is developed to solve this coupled kinetic half-space problems. This allows to obtain a detailed picture of the various interface layers near the nodes and to determine the relevant coefficients in the kinetic derived coupling conditions for the macroscopic equations.Numerical results show the accuracy and efficiency of the approach.

math.AP↗

arXiv2026

On the Generalization of SFT: A Reinforcement Learning Perspective with Reward Rectification

In this work, we present a simple yet theoretically motivated improvement to Supervised Fine-Tuning (SFT) for the Large Language Model (LLM), addressing its limited generalization compared to reinforcement learning (RL). Through mathematical analysis, we reveal that standard SFT gradients implicitly encode a problematic reward structure that may severely restrict the generalization capabilities of model compared to RL. To rectify this, we propose Dynamic Fine-Tuning (\model), stabilizing gradient updates for each token by dynamically rescaling the objective function with the probability of this token. With just a single-line change, the method outperforms standard SFT on multiple difficult benchmarks and base models, from math reasoning to code generation and multi-modal tasks, demonstrating improved generalization. Additionally, \model~achieves competitive results in offline RL settings, providing an effective yet streamlined alternative. By bridging theoretical insights with practical solutions, this work advances the state of SFT. The source code will be available at https://github.com/yongliang-wu/DFT.

cs.LG↗

arXiv2026

Latent Context Compilation: Distilling Long Context into Compact Portable Memory

Efficient long-context LLM deployment is stalled by a dichotomy between amortized compression, which struggles with out-of-distribution generalization, and Test-Time Training, which incurs prohibitive synthetic data costs and requires modifying model weights, creating stateful parameters that complicate concurrent serving. We propose Latent Context Compilation, a framework that fundamentally shifts context processing from adaptation to compilation. By utilizing a disposable LoRA module as a compiler, we distill long contexts into compact buffer tokens -- stateless, portable memory artifacts that are plug-and-play compatible with frozen base models. Crucially, we introduce a self-aligned optimization strategy that eliminates the need for synthetic context-relevant QA pairs. By regularizing context reconstruction task with context-agnostic random queries, we force compressed tokens to reside within the model's existing instruction-following manifold. Experiments with Llama-3.1-8B demonstrate that Latent Context Compilation preserves fine-grained details and reasoning capabilities where prior methods falter, effectively decoupling memory density from model parameters even at a 16x compression ratio.

cs.LG↗

arXiv2025

Interface layers and coupling conditions for discrete kinetic models on networks: a spectral approac

We consider kinetic and related macroscopic equations on networks. A class of linear kinetic BGK models is considered, where the limit equation for small Knudsen numbers is given by the wave equation. Coupling conditions for the macroscopic equations are obtained from the kinetic coupling conditions via an asymptotic analysis near the nodes of the network and the consideration of coupled solutions of kinetic half-space problems. Analytical results are obtained for a discrete velocity version of the coupled half-space problems. Moreover, an efficient spectral method is developed to solve the coupled discrete velocity half-space problems. In particular, this allows to determine the relevant coefficients in the coupling conditions for the macroscopic equations from the underlying kinetic network problem. These coefficients correspond to the so-called extrapolation length for kinetic boundary value problems. Numerical results show the accuracy and fast convergence of the approach. Moreover, a comparison of the kinetic solution on the network with the macroscopic solution is presented.

math.NA↗

arXiv2025

Solving Random Hyperbolic Conservation Laws Using Linear Programming

A novel structure-preserving numerical method to solve random hyperbolic systems of conservation laws is presented. The method uses a concept of generalized, measure-valued solutions to random conservation laws. This yields a linear partial differential equation with respect to the Young measure and allows to compute the approximation based on linear programming problems. We analyze structure-preserving properties of the derived numerical method and discuss its advantages and disadvantages. We numerically demonstrate the approach on the one-dimensional Burgers and isentropic Euler equations and compare with stochastic collocation. In addition, we introduce a discontinuous-flux test in which different entropies used in the linear-program objective select different weak entropy solutions, and we report the corresponding changes in the moments and supports of the Young measure.

math.NA↗

arXiv2025

Turnpike property for hierarchical optimal control problems: from particle systems to hydrodynamic equations

This work is concerned with a hierarchical framework of optimal control problems connecting interacting particle systems, the mean field limit equations, and associated hydrodynamic models. By assuming the existence of solutions, we establish the exponential turnpike property for each level of the hierarchy, showing that optimal trajectories remain close to the associated steady states over long time horizons. The results demonstrate that the exponential turnpike behavior persists consistently across scales, providing a unified connection between microscopic, kinetic, and macroscopic optimal control frameworks.

math.OC↗

arXiv2025

MMAR: Towards Lossless Multi-Modal Auto-Regressive Probabilistic Modeling

Recent advancements in multi-modal large language models have propelled the development of joint probabilistic models capable of both image understanding and generation. However, we have identified that recent methods suffer from loss of image information during understanding task, due to either image discretization or diffusion denoising steps. To address this issue, we propose a novel Multi-Modal Auto-Regressive (MMAR) probabilistic modeling framework. Unlike discretization line of method, MMAR takes in continuous-valued image tokens to avoid information loss in an efficient way. Differing from diffusion-based approaches, we disentangle the diffusion process from auto-regressive backbone model by employing a light-weight diffusion head on top each auto-regressed image patch embedding. In this way, when the model transits from image generation to understanding through text generation, the backbone model's hidden representation of the image is not limited to the last denoising step. To successfully train our method, we also propose a theoretically proven technique that addresses the numerical stability issue and a training strategy that balances the generation and understanding task goals. Extensive evaluations on 18 image understanding benchmarks show that MMAR significantly outperforms most of the existing joint multi-modal models, surpassing the method that employs pre-trained CLIP vision encoder. Meanwhile, MMAR is able to generate high quality images. We also show that our method is scalable with larger data and model size.

cs.CV↗

arXiv2025

Asymptotic-preserving schemes for the initial-boundary value problem of hyperbolic relaxation systems

In this work, we present a numerical method for the initial-boundary value problem (IBVP) of first-order hyperbolic systems with source terms. The scheme directly solves the relaxation system using a relatively coarse mesh and captures the equilibrium behavior quite well, even in the presence of boundary layers. This method extends the concept of asymptotic-preserving schemes from initial-value problems to IBVPs. Moreover, we apply this idea to design a unified numerical scheme for the interface problem of relaxation systems.

math.NA↗

arXiv2025

EE-MLLM: A Data-Efficient and Compute-Efficient Multimodal Large Language Model

Recent advancements in Multimodal Large Language Models (MLLMs) have demonstrated satisfactory performance across various vision-language tasks. Current approaches for vision and language interaction fall into two categories: self-attention-based and cross-attention-based methods. However, both approaches present inherent limitations, forcing a trade-off between data and computational efficiency. To address this issue, we introduce the Data-$\textbf{E}$fficient and Compute-$\textbf{E}$fficient $\textbf{MLLM}$ ($\textbf{EE-MLLM}$). Specifically, we modify the original self-attention mechanism in MLLM to a composite attention mechanism. This mechanism has two key characteristics: 1) eliminating the computational overhead of self-attention among visual tokens to achieve $\textbf{compute efficiency}$, and 2) reusing the weights from each layer of LLM to facilitate effective vision-language modality alignment for $\textbf{data efficiency}$. As a result, EE-MLLM significantly outperforms Flamingo with limited training data, and reduces the prefilling time to 79 ms on an H800 GPU, compared to LLaVA's 277 ms. To further investigate the efficiency of EE-MLLM, we present a training-free variant named EE-MLLM-F, which reduces the computation cost of self-attention-based method without additional training. Experimental results demonstrate the effectiveness of EE-MLLM across a range of benchmarks, including general-purpose datasets like MMBench and SeedBench, as well as fine-grained tasks such as TextVQA and DocVQA.

cs.CV↗

arXiv2025

Slide2Text: Leveraging LLMs for Personalized Textbook Generation from PowerPoint Presentations

The rapid advancements in Large Language Models (LLMs) have revolutionized educational technology, enabling innovative approaches to automated and personalized content creation. This paper introduces Slide2Text, a system that leverages LLMs to transform PowerPoint presentations into customized textbooks. By extracting slide content using OCR, organizing it into a coherent structure, and generating tailored materials such as explanations, exercises, and references, Slide2Text streamlines the textbook creation process. Flexible customization options further enhance its adaptability to diverse educational needs. The system highlights the potential of LLMs in modernizing textbook creation and improving educational accessibility. Future developments will explore multimedia inputs and advanced user customization features.

cs.AI↗

arXiv2025

Number it: Temporal Grounding Videos like Flipping Manga

Video Large Language Models (Vid-LLMs) have made remarkable advancements in comprehending video content for QA dialogue. However, they struggle to extend this visual understanding to tasks requiring precise temporal localization, known as Video Temporal Grounding (VTG). To address this gap, we introduce Number-Prompt (NumPro), a novel method that empowers Vid-LLMs to bridge visual comprehension with temporal grounding by adding unique numerical identifiers to each video frame. Treating a video as a sequence of numbered frame images, NumPro transforms VTG into an intuitive process: flipping through manga panels in sequence. This allows Vid-LLMs to "read" event timelines, accurately linking visual content with corresponding temporal information. Our experiments demonstrate that NumPro significantly boosts VTG performance of top-tier Vid-LLMs without additional computational cost. Furthermore, fine-tuning on a NumPro-enhanced dataset defines a new state-of-the-art for VTG, surpassing previous top-performing methods by up to 6.9\% in mIoU for moment retrieval and 8.5\% in mAP for highlight detection. The code will be available at https://github.com/yongliang-wu/NumPro.

cs.CV↗

arXiv2025

PerturboLLaVA: Reducing Multimodal Hallucinations with Perturbative Visual Training

This paper aims to address the challenge of hallucinations in Multimodal Large Language Models (MLLMs) particularly for dense image captioning tasks. To tackle the challenge, we identify the current lack of a metric that finely measures the caption quality in concept level. We hereby introduce HalFscore, a novel metric built upon the language graph and is designed to evaluate both the accuracy and completeness of dense captions at a granular level. Additionally, we identify the root cause of hallucination as the model's over-reliance on its language prior. To address this, we propose PerturboLLaVA, which reduces the model's reliance on the language prior by incorporating adversarially perturbed text during training. This method enhances the model's focus on visual inputs, effectively reducing hallucinations and producing accurate, image-grounded descriptions without incurring additional computational overhead. PerturboLLaVA significantly improves the fidelity of generated captions, outperforming existing approaches in handling multimodal hallucinations and achieving improved performance across general multimodal benchmarks.

cs.CV↗

arXiv2025

A numerical method for solving the generalized tangent vector of hyperbolic systems

This work is concerned with the computation of the first-order variation for one-dimensional hyperbolic partial differential equations. In the case of shock waves the main challenge is addressed by developing a numerical method to compute the evolution of the generalized tangent vector introduced by Bressan and Marson (1995). Our basic strategy is to combine the conservative numerical schemes and a novel expression of the interface conditions for the tangent vectors along the discontinuity. Based on this, we propose a simple numerical method to compute the tangent vectors for general hyperbolic systems. Numerical results are presented for Burgers' equation and a 2 x 2 hyperbolic system with two genuinely nonlinear fields.

math.NA↗

arXiv2025

A Study on Educational Data Analysis and Personalized Feedback Report Generation Based on Tags and ChatGPT

This study introduces a novel method that employs tag annotation coupled with the ChatGPT language model to analyze student learning behaviors and generate personalized feedback. Central to this approach is the conversion of complex student data into an extensive set of tags, which are then decoded through tailored prompts to deliver constructive feedback that encourages rather than discourages students. This methodology focuses on accurately feeding student data into large language models and crafting prompts that enhance the constructive nature of feedback. The effectiveness of this approach was validated through surveys conducted with over 20 mathematics teachers, who confirmed the reliability of the generated reports. This method can be seamlessly integrated into intelligent adaptive learning systems or provided as a tool to significantly reduce the workload of teachers, providing accurate and timely feedback to students. By transforming raw educational data into interpretable tags, this method supports the provision of efficient and timely personalized learning feedback that offers constructive suggestions tailored to individual learner needs.

cs.AI↗