arXiv subjects

Ruizhe Zhang

Publications and source records attributed to Ruizhe Zhang.

At least 19 recordsLinked to original sources

arXiv2026

AgenticRag-R1: Agentic Reinforcement Learning with Stack Memory for Multi-Step Reasoning, Retrieval and Memorizing

Retrieval-Augmented Generation (RAG) improves the factuality of large language models (LLMs), yet existing RAG systems often struggle with complex, multi-step reasoning that requires adaptive retrieval and continuous revision of intermediate contexts. Recent reinforcement learning (RL)-based agentic RAG methods partially alleviate this issue, but typically rely on coarse-grained action spaces and trajectory-level rewards, resulting in weak reward assignment and a bias toward short-horizon, stereotyped reasoning template. To address, we propose AgenticRag-R1, a RL framework that deeply integrates reasoning, retrieval, and memory via a memory stack and fine-grained action space, supported by hierarchical action-aware rewards and an information-aware trajectory rejection strategy to enable effective long-horizon learning. Experiments across a diverse set of multi-hop, open-domain, and agentic reasoning benchmarks, spanning multiple backbone model sizes, demonstrate that AgenticRag-R1 consistently outperforms strong baselines. Moreover, AgenticRag-R1 learns more robust, interpretable, and memory-aware reasoning behaviors, highlighting the effect of fine-grained action modeling and information-aware optimization for long-horizon reasoning. Our code is anonymous available at https://github.com/jiangxinke/Harness-RL/tree/AgenticRAG-R1-Whitebox.

cs.MA↗

arXiv2026

Hit-and-Run Mixes as Fast as the Ball Walk

Let $K\subset\mathbb{R}^n$ be an isotropic convex body. We prove that the hit-and-run walk, started from any $M$-warm distribution, reaches total-variation distance $\varepsilon$ from the uniform distribution on $K$ in $O\!\left(n^2\psi_n^{-2}\log^3(M/\varepsilon)\right)$ steps, where $\psi_n^{-1}$ is the Kannan-Lov\'asz-Simonovits (KLS) constant. Up to logarithmic factors, this matches the best-known warm-start mixing time for the ball walk. Chen and Eldan [Discrete Comput. Geom. 2026] obtained the same $n^2\psi_n^{-2}$ dependence for hit-and-run, but with polynomial dependence on $M/\varepsilon$. Our result improves that polynomial dependence to a polylogarithmic one, fully resolving their open question about warm-start mixing of hit-and-run in isotropic convex bodies.

cs.DS↗

arXiv2026

Optimal Extrapolation Bounds for Sparse Fourier Sums

We prove an optimal extrapolation theorem for $k$-sparse Fourier sums over arbitrary real frequencies, without any separation assumption, bounding how large such a sum can be just outside an interval on which its energy is observed. For every $g(t)=\sum_{j=1}^k v_j e^{i\lambda_jt}$ with $\lambda_j\in\mathbb R$ and every $x\ge1$, $$ |g(x)|\le k^{O(1)}\exp(O(k\mathop{\mathrm{arcosh}} x))\|g\|_{L^2[-1,1]} . $$ In the endpoint regime, this refines to the explicit bound $$ |g(1+\delta)|\le O(k)\exp(O(k\sqrt\delta))\|g\|_{L^2[-1,1]}, \qquad 0\le\delta\le1 . $$ This improves on the $\exp(O(k^2\log k\cdot\delta))$ growth estimate of Chen and Price (ICALP 2019), and the exponential scaling is optimal up to constants and polynomial factors in $k$. As an algorithmic consequence, we improve the cluster-center resolution of Chen--Price's clustered-frequency recovery algorithm by a factor of $k$, while preserving its sample complexity up to logarithmic factors. We also obtain exterior leverage-score and transfer bounds for sparse Fourier feature spaces, converting in-domain active-regression guarantees into essentially sharp prediction guarantees just outside the sampling interval.

cs.DS↗

arXiv2026

ScaffoldAgent: Utility-Guided Dynamic Outline Optimization for Open-Ended Deep Research

Open-ended deep research (OEDR) requires systems to acquire knowledge through multi-round retrieval and generate coherent long-form reports. The outline plays a central role as a structural scaffold that coordinates retrieval, evidence organization, and generation. However, existing methods either fix the outline before writing or refine it with local heuristics, leading to scaffold drift under continuous information accumulation and delayed feedback for evaluating outline modifications. We propose ScaffoldAgent, a utility-guided dynamic outline optimization framework for OEDR. ScaffoldAgent models outline evolution as a structured decision process with three operations: Expansion, Contraction, and Revision, enabling controlled updates to the report scaffold. It further introduces a utility-guided feedback mechanism that estimates the downstream value of each outline operation from retrieval gain, structural coherence, and trial-generation quality. The resulting utility signal guides node selection, operation scheduling, and termination during inference. Experiments on DeepResearch Bench and DeepResearch Gym show that ScaffoldAgent consistently improves long-form report generation and factual grounding over existing deep research agents.

cs.AI↗

arXiv2026

Diff-Instruct with Diffused Reward: Towards Principled One-step Generator RL

Recent advances in one-step text-to-image generation have enabled real-time synthesis with remarkable efficiency and quality. Previous reinforcement learning methods for one-step generators combine image-space reward optimization with diffusion noisy-space distribution matching. This paradigm brings challenges due to a mismatch between terminal reward optimization and the underlying generative dynamics. As a result, optimization tends to exploit stochastic degrees of freedom, often improving reward at the expense of image fidelity. To address this issue, we propose Diff-Instruct with Diffused Reward (DIDR), a data-free trajectory-level alignment framework derived from Integral KL minimization. DIDR propagates the RLHF-optimal reward-tilted clean-image distribution across all noise levels along the diffusion trajectory. We show that this objective admits the same minimizer as clean-image RLHF, while naturally inducing the Diffused Reward Score (DRS), which acts as a reward-driven correction to the reference score function. To make this practical, we further introduce the Diffused Reward Proxy (DRP), an efficient estimator of DRS based on differentiable short-step denoising. Extensive experiments demonstrate that DIDR consistently Pareto-dominates existing one-step SDXL baselines. Moreover, when transferred to a 6B DiT backbone (Z-Image), DIDR surpasses its 50-step teacher in preference alignment while requiring only a single generation step.

cs.CV↗

arXiv2026

Overcoming the Lamb Shift in System-Bath Interaction Models via KMS Detailed Balance: High-Accuracy Thermalization with Time-Bounded Interactions

We investigate quantum thermal state preparation algorithms based on system-bath interactions and uncover a surprising phenomenon in the weak-coupling regime. We rigorously prove that, if the system-bath interaction is engineered so that the transition part of the approximate Lindbladian generator satisfies the Kubo--Martin--Schwinger (KMS) detailed balance condition, then the unique fixed point of the dynamics can be made arbitrarily close to the Gibbs state in the weak-coupling limit, regardless of the structure of the Lamb shift term. Importantly, this remains true even when the approximate Lindbladian differs substantially from the ideal Davies generator and the Lamb shift term does not commute with the thermal state. Our result shows that the role of the KMS detailed balance condition extends well beyond standard Lindbladian dynamics, serving as a general principle for a broader class of dissipative systems. Furthermore, by combining this with a general perturbation framework, we bound the mixing time of the dynamics and establish an end-to-end complexity of $\mathcal{O}(\varepsilon^{-1})$ for Gibbs state preparation. These guarantees apply to any Hamiltonian whose associated KMS-detailed-balance Lindbladian is known to be fast mixing.

quant-ph↗

arXiv2026

Quantum Filtering and Analysis of Multiplicities in Eigenvalue Spectra

Fine-grained spectral properties of quantum Hamiltonians, including both eigenvalues and their multiplicities, provide useful information for characterizing many-body quantum systems as well as for understanding phenomena such as topological order. Extracting such information with small additive error is $\#\textsf{BQP}$-complete in the worst case. In this work, we introduce QFAMES (Quantum Filtering and Analysis of Multiplicities in Eigenvalue Spectra), a quantum algorithm that efficiently identifies clusters of closely spaced dominant eigenvalues and determines their multiplicities under physically motivated assumptions, which allows us to bypass worst-case complexity barriers. QFAMES also enables the estimation of observable expectation values within targeted energy clusters, providing a powerful tool for studying quantum phase transitions and other physical properties. We validate the effectiveness of QFAMES through numerical demonstrations, including its applications to characterizing quantum phases in the transverse-field Ising model and estimating the ground-state degeneracy of a topologically ordered phase in the two-dimensional toric code model. We also generalize QFAMES to the setting of mixed initial states. Our approach offers rigorous theoretical guarantees and significant advantages over existing subspace-based quantum spectral analysis methods, particularly in terms of the sample complexity and the ability to resolve degeneracies.

quant-ph↗

arXiv2026

InfoMamba: An Attention-Free Hybrid Mamba-Transformer Model

Balancing fine-grained local modeling with long-range dependency capture under computational constraints remains a central challenge in sequence modeling. While Transformers provide strong token mixing, they suffer from quadratic complexity, whereas Mamba-style selective state-space models (SSMs) scale linearly but often struggle to capture high-rank and synchronous global interactions. We present a consistency boundary analysis that characterizes when diagonal short-memory SSMs can approximate causal attention and identifies structural gaps that remain. Motivated by this analysis, we propose InfoMamba, an attention-free hybrid architecture. InfoMamba replaces token-level self-attention with a concept bottleneck linear filtering layer that serves as a minimal-bandwidth global interface and integrates it with a selective recurrent stream through information-maximizing fusion (IMF). IMF dynamically injects global context into the SSM dynamics and encourages complementary information usage through a mutual-information-inspired objective. Extensive experiments on classification, dense prediction, and non-vision tasks show that InfoMamba consistently outperforms strong Transformer and SSM baselines, achieving competitive accuracy-efficiency trade-offs while maintaining near-linear scaling.

cs.LG↗

arXiv2026

Beyond Classical Attention: Quantum Attention for Scalable Computation

As large language models (LLMs) demonstrate outstanding performance across various tasks, attention-driven models have profoundly transformed the field of machine learning. Since attention computations account for the primary computational overhead in both model inference and training, efficiently computing attention matrices has become one of the core challenges in accelerating large language models. It is well-known that quantum machines possess computational advantages over classical machines, and the role of quantum computing in LLMs remains largely unexplored. In this work, we focus on leveraging the Grover search algorithm to efficiently compute a sparse attention matrix. Through comparisons with classical algorithms, we demonstrate that our method achieves quantum acceleration in polynomial time. Additionally, we observe that the generated quantum attention matrices naturally exhibit low-rank structures, providing further theoretical support for efficient modeling. Moreover, within the specific context of attention matrix computation, we conduct a systematic and detailed analysis of the error and time complexity of the proposed algorithm.

quant-ph↗

arXiv2026

Sampling via Stochastic Interpolants by Langevin-based Velocity and Initialization Estimation in Flow ODEs

We propose a novel method for sampling from unnormalized Boltzmann densities based on a probability flow ordinary differential equation (ODE) derived from linear stochastic interpolants. The key innovation of our approach is the use of a sequence of Langevin samplers to enable efficient simulation of the flow. Specifically, these Langevin samplers are employed (i) to generate samples from the interpolant distribution at intermediate times and (ii) to construct, starting from these intermediate times, a robust estimator of the velocity field governing the probability flow ODE. Theoretically, we provide convergence guarantees for both Langevin components, and establish a non-asymptotic convergence rate for the probability flow ODE. Extensive numerical experiments demonstrate the efficiency of the proposed method on challenging multimodal distributions across a range of dimensions, as well as its effectiveness in Bayesian inference tasks.

math.NA↗

arXiv2026

StackPlanner: A Centralized Hierarchical Multi-Agent System with Task-Experience Memory Management

Multi-agent systems based on large language models, particularly centralized architectures, have recently shown strong potential for complex and knowledge-intensive tasks. However, central agents often suffer from unstable long-horizon collaboration due to the lack of memory management, leading to context bloat, error accumulation, and poor cross-task generalization. To address both task-level memory inefficiency and the inability to reuse coordination experience, we propose StackPlanner, a hierarchical multi-agent framework with explicit memory control. StackPlanner addresses these challenges by decoupling high-level coordination from subtask execution with active task-level memory control, and by learning to retrieve and exploit reusable coordination experience via structured experience memory and reinforcement learning. Experiments on multiple deep-search and agent system benchmarks demonstrate the effectiveness of our approach in enabling reliable long-horizon multi-agent collaboration.

cs.AI↗

arXiv2025

Beyond the ACE Score: Replicable Combinations of Adverse Childhood Experiences That Worsen Depression Risk

Adverse childhood experiences (ACEs) are categories of childhood abuse, neglect, and household dysfunction. Screening by a single additive ACE score (e.g., a $\ge 4$ cutoff) has poor individual-level discrimination. We instead identify replicable combinations of ACEs that elevate adult depression risk. Our data turnover framework enables a single research team to explore, confirm, and replicate within one observational dataset while controlling the family-wise error rate. We integrate isotonic subgroup selection (ISS) to estimate a higher-risk subgroup under a monotonicity assumption -- additional ACE exposure or higher intensity cannot reduce depression risk. We pre-specify a risk threshold $τ$ corresponding to roughly a two-fold increase in the odds of depression relative to the no-ACE baseline. Within data turnover, the prespecified component improves power while maintaining FWER control, as demonstrated in simulations. Guided by EDA, we adopt frequency coding for ACE items, retaining intensity information that reduces false positives relative to binary or score codings. The result is a replicable, pattern-based higher-risk subgroup. On held-out BRFSS 2022, we show that, at the same level of specificity (0.95), using our replicable subgroup as the screening rule increases sensitivity by 26\% compared with an ACE-score cutoff, yielding concrete triggers that are straightforward to implement and help target scarce clinical screening resources toward truly higher-risk profiles.

stat.AP↗

arXiv2025

Protocol For An Observational Study On The Effects Of Combinations Of Adverse Childhood Experiences On Adult Depression

Adverse childhood experiences (ACEs) have been linked to a wide range of negative health outcomes in adulthood. However, few studies have investigated what specific combinations of ACEs most substantially impact mental health. In this article, we provide the protocol for our observational study of the effects of combinations of ACEs on adult depression. We use data from the 2023 Behavioral Risk Factor Surveillance System (BRFSS) to assess these effects. We will evaluate the replicability of our findings by splitting the sample into two discrete subpopulations of individuals. We employ data turnover for this analysis, enabling a single team of statisticians and domain experts to collaboratively evaluate the strength of evidence, and also integrating both qualitative and quantitative insights from exploratory data analysis. We outline our analysis plan using this method and conclude with a brief discussion of several specifics for our study.

stat.AP↗

arXiv2025

Low-Rank Tensor Recovery via Variational Schatten-p Quasi-Norm and Jacobian Regularization

Higher-order tensors are well-suited for representing multi-dimensional data, such as images and videos, which typically characterize low-rank structures. Low-rank tensor decomposition has become essential in machine learning and computer vision, but existing methods like Tucker decomposition offer flexibility at the expense of interpretability. The CANDECOMP/PARAFAC (CP) decomposition provides a natural and interpretable structure, while obtaining a sparse solutions remains challenging. Leveraging the rich properties of CP decomposition, we propose a CP-based low-rank tensor function parameterized by neural networks (NN) for implicit neural representation. This approach can model the tensor both on-grid and beyond grid, fully utilizing the non-linearity of NN with theoretical guarantees on excess risk bounds. To achieve sparser CP decomposition, we introduce a variational Schatten-p quasi-norm to prune redundant rank-1 components and prove that it serves as a common upper bound for the Schatten-p quasi-norms of arbitrary unfolding matrices. For smoothness, we propose a regularization term based on the spectral norm of the Jacobian and Hutchinson's trace estimator. The proposed smoothness regularization is SVD-free and avoids explicit chain rule derivations. It can serve as an alternative to Total Variation (TV) regularization in image denoising tasks and is naturally applicable to implicit neural representation. Extensive experiments on multi-dimensional data recovery tasks, including image inpainting, denoising, and point cloud upsampling, demonstrate the superiority and versatility of our method compared to state-of-the-art approaches. The code is available at https://github.com/CZY-Code/CP-Pruner.

cs.CV↗

arXiv2025

Parenting: Optimizing Knowledge Selection of Retrieval-Augmented Language Models with Parameter Decoupling and Tailored Tuning

Retrieval-Augmented Generation (RAG) offers an effective solution to the issues faced by Large Language Models (LLMs) in hallucination generation and knowledge obsolescence by incorporating externally retrieved knowledge. However, existing methods lack effective control mechanisms for integrating internal and external knowledge. Inspired by human cognitive processes, we propose Parenting, a novel framework that decouples, identifies, and purposefully optimizes parameter subspaces related to adherence and robustness. Specifically, Parenting utilizes a key parameter mining method that combines forward and backward propagation signals to localize subspaces representing different capabilities. Then, Parenting employs a type-tailored tuning strategy, applying specific and appropriate optimizations to different subspaces, aiming to achieve a balanced enhancement of both adherence and robustness. Extensive experiments on various datasets and models validate the effectiveness and generalizability of our method.

cs.CL↗

arXiv2025

EL4NER: Ensemble Learning for Named Entity Recognition via Multiple Small-Parameter Large Language Models

In-Context Learning (ICL) technique based on Large Language Models (LLMs) has gained prominence in Named Entity Recognition (NER) tasks for its lower computing resource consumption, less manual labeling overhead, and stronger generalizability. Nevertheless, most ICL-based NER methods depend on large-parameter LLMs: the open-source models demand substantial computational resources for deployment and inference, while the closed-source ones incur high API costs, raise data-privacy concerns, and hinder community collaboration. To address this question, we propose an Ensemble Learning Method for Named Entity Recognition (EL4NER), which aims at aggregating the ICL outputs of multiple open-source, small-parameter LLMs to enhance overall performance in NER tasks at less deployment and inference cost. Specifically, our method comprises three key components. First, we design a task decomposition-based pipeline that facilitates deep, multi-stage ensemble learning. Second, we introduce a novel span-level sentence similarity algorithm to establish an ICL demonstration retrieval mechanism better suited for NER tasks. Third, we incorporate a self-validation mechanism to mitigate the noise introduced during the ensemble process. We evaluated EL4NER on multiple widely adopted NER datasets from diverse domains. Our experimental results indicate that EL4NER surpasses most closed-source, large-parameter LLM-based methods at a lower parameter cost and even attains state-of-the-art (SOTA) performance among ICL-based methods on certain datasets. These results show the parameter efficiency of EL4NER and underscore the feasibility of employing open-source, small-parameter LLMs within the ICL paradigm for NER tasks.

cs.CL↗

arXiv2025

Polynomial-Time Preparation of Low-Temperature Gibbs States for 2D Toric Code

We propose a polynomial-time algorithm for preparing the Gibbs state of the two-dimensional toric code Hamiltonian at any temperature, starting from any initial condition, significantly improving upon prior estimates that suggested exponential scaling with inverse temperature. Our approach combines the Lindblad dynamics using a local Davies generator with simple global jump operators to enable efficient transitions between logical sectors. Our proof also shows that the Lindblad dynamics with a digitally implemented low-temperature local Davies generator is able to efficiently drive the quantum state towards the ground state manifold.

quant-ph↗

arXiv2024

RAGraph: A General Retrieval-Augmented Graph Learning Framework

Graph Neural Networks (GNNs) have become essential in interpreting relational data across various domains, yet, they often struggle to generalize to unseen graph data that differs markedly from training instances. In this paper, we introduce a novel framework called General Retrieval-Augmented Graph Learning (RAGraph), which brings external graph data into the general graph foundation model to improve model generalization on unseen scenarios. On the top of our framework is a toy graph vector library that we established, which captures key attributes, such as features and task-specific label information. During inference, the RAGraph adeptly retrieves similar toy graphs based on key similarities in downstream tasks, integrating the retrieved data to enrich the learning context via the message-passing prompting mechanism. Our extensive experimental evaluations demonstrate that RAGraph significantly outperforms state-of-the-art graph learning methods in multiple tasks such as node classification, link prediction, and graph classification across both dynamic and static datasets. Furthermore, extensive testing confirms that RAGraph consistently maintains high performance without the need for task-specific fine-tuning, highlighting its adaptability, robustness, and broad applicability.

cs.LG↗