arXiv subjects

Mengqi Zhang

Publications and source records attributed to Mengqi Zhang.

At least 37 records · Page 2Linked to original sources

arXiv2026

Style-Instructed Mask-Free Virtual Try On

Virtual Try-On is a promising research area with broad applications in e-commerce and everyday life, enabling users to visualize garments on themselves or others before purchase. Most existing methods depend on predefined or user-specified masks to guide garment placement, but their performance is highly sensitive to mask quality, often causing misalignment or artifacts, and introduces redundant steps for users. To overcome these limitations, we propose a mask-free virtual try-on framework that requires only minimal modifications to the underlying architecture while remaining compatible with common diffusion-based pipelines. To address the increased ambiguity in the absence of masks, we integrate an attention-based guidance mechanism that explicitly directs the model to focus on the target garment region and improves correspondence between the garment and the person. Additionally, we incorporate instruction prompts, allowing users to flexibly control garment categories and wearing styles, addressing the underutilization of prompts in prior work and improving interaction flexibility. Both qualitative and quantitative evaluations across multiple datasets demonstrate that our approach consistently outperforms existing methods, producing more accurate, robust, and user-friendly try-on results.

cs.GR↗

arXiv2026

DiffuGR: Generative Document Retrieval with Diffusion Language Models

Generative retrieval (GR) reframes document retrieval as an end-to-end task of generating sequential document identifiers (DocIDs). Existing GR methods predominantly rely on left-to-right auto-regressive decoding, which suffers from two fundamental limitations: (i) a \emph{mismatch between DocID generation and natural language generation}, whereby an incorrect DocID token generated at an early step can lead to entirely erroneous retrieval; and (ii) an \emph{inability to dynamically balance the trade-off between retrieval efficiency and accuracy}, which is crucial for practical applications. To tackle these challenges, we propose generative document retrieval with diffusion language models, termed \emph{DiffuGR}. DiffuGR formulates DocID generation as a discrete diffusion process. During training, DocIDs are corrupted through a stochastic masking process, and a diffusion language model is trained to recover them under a retrieval-aware objective. For inference, DiffuGR generates DocID tokens in parallel and refines them through a controllable number of denoising steps. Unlike auto-regressive decoding, DiffuGR introduce \emph{a novel mechanism to first generate plenty of confident DocID tokens and then refine the generation through diffusion-based denoising}. Moreover, DiffuGR also offers \emph{explicit runtime control over the quality-latency tradeoff}. Extensive experiments on widely-applied retrieval benchmarks show that DiffuGR outperforms strong auto-regressive generative retrievers. Additionally, we verify that DiffuGR achieves flexible control over the quality-latency trade-off via variable denoising budgets.

cs.IR↗

arXiv2025

Mixture-of-RAG: Integrating Text and Tables with Large Language Models

Large language models (LLMs) achieve optimal utility when their responses are grounded in external knowledge sources. However, real-world documents, such as annual reports, scientific papers, and clinical guidelines, frequently combine extensive narrative content with complex, hierarchically structured tables. While existing retrieval-augmented generation (RAG) systems effectively integrate LLMs' generative capabilities with external retrieval-based information, their performance significantly deteriorates especially processing such heterogeneous text-table hierarchies. To address this limitation, we formalize the task of Heterogeneous Document RAG, which requires joint retrieval and reasoning across textual and hierarchical tabular data. We propose MixRAG, a novel three-stage framework: (i) hierarchy row-and-column-level (H-RCL) representation that preserves hierarchical structure and heterogeneous relationship, (ii) an ensemble retriever with LLM-based reranking for evidence alignment, and (iii) multi-step reasoning decomposition via a RECAP prompt strategy. To bridge the gap in available data for this domain, we release the dataset DocRAGLib, a 2k-document corpus paired with automatically aligned text-table summaries and gold document annotations. The comprehensive experiment results demonstrate that MixRAG boosts top-1 retrieval by 46% over strong text-only, table-only, and naive-mixture baselines, establishing new state-of-the-art performance for mixed-modality document grounding.

cs.IR↗

arXiv2025

Topological Structure Learning Should Be A Research Priority for LLM-Based Multi-Agent Systems

Large Language Model-based Multi-Agent Systems (MASs) have emerged as a powerful paradigm for tackling complex tasks through collaborative intelligence. However, the topology of these systems--how agents in MASs should be configured, connected, and coordinated--remains largely unexplored. In this position paper, we call for a paradigm shift toward \emph{topology-aware MASs} that explicitly model and dynamically optimize the structure of inter-agent interactions. We identify three fundamental components--agents, communication links, and overall topology--that collectively determine the system's adaptability, efficiency, robustness, and fairness. To operationalize this vision, we introduce a systematic three-stage framework: 1) agent selection, 2) structure profiling, and 3) topology synthesis. This framework not only provides a principled foundation for designing MASs but also opens new research frontiers across language modeling, reinforcement learning, graph learning, and generative modeling to ultimately unleash their full potential in complex real-world applications. We conclude by outlining key challenges and opportunities in MASs evaluation. We hope our framework and perspectives offer critical new insights in the era of agentic AI.

cs.MA↗

arXiv2025

On the intracyclic instability in Stokes layers

Time-dependent fluid dynamics plays a crucial role in both natural phenomena and industrial applications. Understanding the flow instabilities and transitions within these dynamical systems is essential for predicting and controlling their unsteady behaviour. A classic example of time-dependent flow is the Stokes layer. To study the transition mechanism in this flow, we employ the Finite-Time Lyapunov Exponent (FTLE) to demonstrate that a linear energy amplification mechanism may explain the intracyclic instability in the transitional Stokes layer, supported by favourable comparisons with experimental measurements of axial turbulence intensity. This complements existing theories applied to the Stokes layer in the literature, including the Floquet analysis and the instantaneous/momentary analyses, which have struggled to capture this experimental observation accurately. The FTLE analysis is closely related to the transient growth analysis, formulated as an optimisation problem of the disturbance energy growth over time. We found that the energy amplification weakens as the finite Stokes layer becomes more confined and the oscillating frequency has a non-monotonic effect on the maximum transient growth. Based on these results, we recommend future experimental studies to validate this linear mechanism.

physics.flu-dyn↗

arXiv2025

Uncovering Overfitting in Large Language Model Editing

Knowledge editing has been proposed as an effective method for updating and correcting the internal knowledge of Large Language Models (LLMs). However, existing editing methods often struggle with complex tasks, such as multi-hop reasoning. In this paper, we identify and investigate the phenomenon of Editing Overfit, where edited models assign disproportionately high probabilities to the edit target, hindering the generalization of new knowledge in complex scenarios. We attribute this issue to the current editing paradigm, which places excessive emphasis on the direct correspondence between the input prompt and the edit target for each edit sample. To further explore this issue, we introduce a new benchmark, EVOKE (EValuation of Editing Overfit in Knowledge Editing), along with fine-grained evaluation metrics. Through comprehensive experiments and analysis, we demonstrate that Editing Overfit is prevalent in current editing methods and that common overfitting mitigation strategies are ineffective in knowledge editing. To overcome this, inspired by LLMs' knowledge recall mechanisms, we propose a new plug-and-play strategy called Learn the Inference (LTI), which introduce a Multi-stage Inference Constraint module to guide the edited models in recalling new knowledge similarly to how unedited LLMs leverage knowledge through in-context learning. Extensive experimental results across a wide range of tasks validate the effectiveness of LTI in mitigating Editing Overfit.

cs.CL↗

arXiv2025

A Diffuse-Interface Marangoni Instability

We investigate a novel Marangoni-induced instability that arises exclusively in diffuse fluid interfaces, absent in classical sharp-interface models. Using a validated phase-field Navier-Stokes-Allen-Cahn framework, we linearize the governing equations to analyze the onset and development of interfacial instability driven by solute-induced surface tension gradients. A critical interfacial thickness scaling inversely with the Marangoni number, $δ_\mathrm{cr} \sim Ma^{-1}$, emerges from the balance between advective and diffusive transport. Unlike sharp-interface scenarios where matched viscosity and diffusivity stabilize the interface, finite thickness induces asymmetric solute distributions and tangential velocity shifts that destabilize the system. We identify universal power-law scalings of velocity and concentration offsets with a modified Marangoni number $Ma^δ$, independent of capillary number and interfacial mobility. A critical crossover at $Ma^δ\approx 590$ distinguishes diffusion-dominated stabilization from advection-driven destabilization. These findings highlight the importance of diffuse-interface effects in multiphase flows, with implications for miscible fluids, soft matter, and microfluidics where interfacial thickness and coupled transport phenomena are non-negligible.

physics.flu-dyn↗

arXiv2025

LLM Unlearning Should Be Form-Independent

Large Language Model (LLM) unlearning aims to erase or suppress undesirable knowledge within the model, offering promise for controlling harmful or private information to prevent misuse. However, recent studies highlight its limited efficacy in real-world scenarios, hindering practical adoption. In this study, we identify a pervasive issue underlying many downstream failures: the effectiveness of existing unlearning methods heavily depends on the form of training samples and frequently fails to generalize to alternate expressions of the same knowledge. We formally characterize this problem as Form-Dependent Bias and systematically investigate its specific manifestation patterns across various downstream tasks. To quantify its prevalence and support future research, we introduce ORT, a novel benchmark designed to evaluate the robustness of unlearning methods against variations in knowledge expression. Results reveal that Form-Dependent Bias is both widespread and severe among current techniques. We argue that LLM unlearning should be form-independent to address the endless forms of downstream tasks encountered in real-world security-critical scenarios. Towards this goal, we introduce Rank-one Concept Redirection (ROCR), a novel training-free method, as a promising solution path. ROCR performs unlearning by targeting the invariants in downstream tasks, specifically the activated dangerous concepts. It is capable of modifying model parameters within seconds to redirect the model's perception of a specific unlearning target concept to another harmless concept. Extensive experiments demonstrate that ROCR significantly improves unlearning effectiveness compared to traditional methods while generating highly natural outputs.

cs.CL↗

arXiv2025

MELoRA: Mini-Ensemble Low-Rank Adapters for Parameter-Efficient Fine-Tuning

Parameter-efficient fine-tuning (PEFT) is a popular method for tailoring pre-trained large language models (LLMs), especially as the models' scale and the diversity of tasks increase. Low-rank adaptation (LoRA) is based on the idea that the adaptation process is intrinsically low-dimensional, i.e., significant model changes can be represented with relatively few parameters. However, decreasing the rank encounters challenges with generalization errors for specific tasks when compared to full-parameter fine-tuning. We present MELoRA, a mini-ensemble low-rank adapters that uses fewer trainable parameters while maintaining a higher rank, thereby offering improved performance potential. The core idea is to freeze original pretrained weights and train a group of mini LoRAs with only a small number of parameters. This can capture a significant degree of diversity among mini LoRAs, thus promoting better generalization ability. We conduct a theoretical analysis and empirical studies on various NLP tasks. Our experimental results show that, compared to LoRA, MELoRA achieves better performance with 8 times fewer trainable parameters on natural language understanding tasks and 36 times fewer trainable parameters on instruction following tasks, which demonstrates the effectiveness of MELoRA.

cs.CL↗

arXiv2025

Reinforcement-learning-assisted control of four-roll mills: geometric symmetry and inertial effect

Embedding the intrinsic symmetry of a flow system in training its machine learning algorithms has become a significant trend in the recent surge of their application in fluid mechanics. This paper leverages the geometric symmetry of a four-roll mill (FRM) to enhance its training efficiency. Stabilizing and precisely controlling droplet trajectories in a FRM is challenging due to the unstable nature of the extensional flow with a saddle point. Extending the work of Vona & Lauga, this study applies Deep Reinforcement Learning (DRL) to effectively guide a displaced droplet to the center of the FRM. Through direct numerical simulations, we explore the applicability of DRL in controlling FRM flow with moderate inertial effects, i.e., Reynolds number $\sim\mathcal{O}(1)$, a nonlinear regime previously unexplored. The FRM's geometric symmetry allows control policies trained in one of the eight sub-quadrants to be extended to the entire domain, reducing training costs. Our results indicate that the DRL-based control method can successfully guide a displaced droplet to the target center with robust performance across various starting positions, even from substantially far distances. The work also highlights potential directions for future research, particularly focusing on efficiently addressing the delay effects in flow response caused by inertia. This study presents new advances in controlling droplet trajectories in more nonlinear and complex situations, with potential applications to other nonlinear flows. The geometric symmetry used in this cutting-edge reinforcement learning approach can also be applied to other control methods.

physics.flu-dyn↗

arXiv2025

Constrained Auto-Regressive Decoding Constrains Generative Retrieval

Generative retrieval seeks to replace traditional search index data structures with a single large-scale neural network, offering the potential for improved efficiency and seamless integration with generative large language models. As an end-to-end paradigm, generative retrieval adopts a learned differentiable search index to conduct retrieval by directly generating document identifiers through corpus-specific constrained decoding. The generalization capabilities of generative retrieval on out-of-distribution corpora have gathered significant attention. In this paper, we examine the inherent limitations of constrained auto-regressive generation from two essential perspectives: constraints and beam search. We begin with the Bayes-optimal setting where the generative retrieval model exactly captures the underlying relevance distribution of all possible documents. Then we apply the model to specific corpora by simply adding corpus-specific constraints. Our main findings are two-fold: (i) For the effect of constraints, we derive a lower bound of the error, in terms of the KL divergence between the ground-truth and the model-predicted step-wise marginal distributions. (ii) For the beam search algorithm used during generation, we reveal that the usage of marginal distributions may not be an ideal approach. This paper aims to improve our theoretical understanding of the generalization capabilities of the auto-regressive decoding retrieval paradigm, laying a foundation for its limitations and inspiring future advancements toward more robust and generalizable generative retrieval.

cs.IR↗

arXiv2025

Open Problems and a Hypothetical Path Forward in LLM Knowledge Paradigms

Knowledge is fundamental to the overall capabilities of Large Language Models (LLMs). The knowledge paradigm of a model, which dictates how it encodes and utilizes knowledge, significantly affects its performance. Despite the continuous development of LLMs under existing knowledge paradigms, issues within these frameworks continue to constrain model potential. This blog post highlight three critical open problems limiting model capabilities: (1) challenges in knowledge updating for LLMs, (2) the failure of reverse knowledge generalization (the reversal curse), and (3) conflicts in internal knowledge. We review recent progress made in addressing these issues and discuss potential general solutions. Based on observations in these areas, we propose a hypothetical paradigm based on Contextual Knowledge Scaling, and further outline implementation pathways that remain feasible within contemporary techniques. Evidence suggests this approach holds potential to address current shortcomings, serving as our vision for future model paradigms. This blog post aims to provide researchers with a brief overview of progress in LLM knowledge systems, while provide inspiration for the development of next-generation model architectures.

cs.CL↗

arXiv2025

UIPE: Enhancing LLM Unlearning by Removing Knowledge Related to Forgetting Targets

Large Language Models (LLMs) inevitably acquire harmful information during training on massive datasets. LLM unlearning aims to eliminate the influence of such harmful information while maintaining the model's overall performance. Existing unlearning methods, represented by gradient ascent-based approaches, primarily focus on forgetting target data while overlooking the crucial impact of logically related knowledge on the effectiveness of unlearning. In this paper, through both theoretical and experimental analyses, we first demonstrate that a key reason for the suboptimal unlearning performance is that models can reconstruct the target content through reasoning with logically related knowledge. To address this issue, we propose Unlearning Improvement via Parameter Extrapolation (UIPE), a method that removes knowledge highly correlated with the forgetting targets. Experimental results show that UIPE significantly enhances the performance of various mainstream LLM unlearning methods on the TOFU benchmark.

cs.CL↗

arXiv2025

Reinforcement-learning-based control of turbulent channel flows at high Reynolds numbers

Deep reinforcement learning (DRL) is employed to develop control strategies for drag reduction in direct numerical simulations (DNS) of turbulent channel flows at high Reynolds numbers. The DRL agent uses near-wall streamwise velocity fluctuations as input to modulate wall blowing and suction velocities. These DRL-based strategies achieve significant drag reduction, with maximum rates of 35.6% at Re_τ=180, 30.4% at Re_τ=550, and 27.7% at Re_τ=1000, outperforming traditional opposition control methods. Expanded range of wall actions further enhances drag reduction, although effectiveness decreases at higher Reynolds numbers. The DRL models elevate the virtual wall through blowing and suction, aiding in drag reduction. However, at higher Reynolds numbers, the amplitude modulation of large-scale structures significantly increases the residual Reynolds stress on the virtual wall, diminishing the drag reduction. Analysis of budget equations provides a systematic understanding of the drag reduction dynamics behind. DRL models reduce skin friction by inhibiting the redistribution of wall-normal turbulent kinetic energy. This further suppresses the wall-normal velocity fluctuations, reducing the production of Reynolds stress, thereby decreasing skin friction. This study showcases the successful application of DRL in turbulence control at high Reynolds numbers and elucidates the nonlinear control mechanisms underlying the observed drag reduction.

physics.flu-dyn↗

arXiv2024

Numerical solutions of fixed points in two-dimensional Kuramoto-Sivashinsky equation expedited by reinforcement learning

This paper presents a combined approach to enhancing the effectiveness of Jacobian-Free Newton-Krylov (JFNK) method by deep reinforcement learning (DRL) in identifying fixed points within the 2D Kuramoto-Sivashinsky Equation (KSE). JFNK approach entails a good initial guess for improved convergence when searching for fixed points. With a properly defined reward function, we utilise DRL as a preliminary step to enhance the initial guess in the converging process. We report new results of fixed points in the 2D KSE which have not been reported in the literature. Additionally, we explored control optimization for the 2D KSE to navigate the system trajectories between known fixed points, based on parallel reinforcement learning techniques. This combined method underscores the improved JFNK approach to finding new fixed-point solutions within the context of 2D KSE, which may be instructive for other high-dimensional dynamical systems.

cs.LG↗

arXiv2024

ASLoRA: Adaptive Sharing Low-Rank Adaptation Across Layers

As large language models (LLMs) grow in size, traditional full fine-tuning becomes increasingly impractical due to its high computational and storage costs. Although popular parameter-efficient fine-tuning methods, such as LoRA, have significantly reduced the number of tunable parameters, there is still room for further optimization. In this work, we propose ASLoRA, a cross-layer parameter-sharing strategy combining global sharing with partial adaptive sharing. Specifically, we share the low-rank matrix A across all layers and adaptively merge matrix B during training. This sharing mechanism not only mitigates overfitting effectively but also captures inter-layer dependencies, significantly enhancing the model's representational capability. We conduct extensive experiments on various NLP tasks, showing that ASLoRA outperforms LoRA while using less than 25% of the parameters, highlighting its flexibility and superior parameter efficiency. Furthermore, in-depth analyses of the adaptive sharing strategy confirm its significant advantages in enhancing both model flexibility and task adaptability.

cs.CL↗

arXiv2024

Marangoni Interfacial Instability Induced by Solute Transfer Across Liquid-Liquid Interfaces

This study presents analytical and numerical investigations of Marangoni interfacial instability in a two-liquid-layer system with constant solute transfer across the interface. While previous research has established that both diffusivity and viscosity ratios affect hydrodynamic stability via the Marangoni effect, the specific nonlinear dynamics and the role of interfacial deformation remain fully unclear. To address this, we developed a phase-field-based numerical model, validated against linear stability analysis and existing theories. The validated parameter space includes Schmidt number, Marangoni number, Capillary number, and the diffusivity and viscosity ratio between the two layers. Our finding shows that solute transfer from a less diffusive layer triggers short-wave instability, governed by the critical Marangoni number, while solute transfer into a less viscous layer induces long-wave instability, controlled by the critical Capillary number. Nonlinear simulations reveal distinct field coupling behaviors: in the diffusivity-ratio-driven instability, the spatially averaged flow intensity remains symmetric about a flat interface, while solute gradient is uneven. In contrast, in viscosity-ratio-driven instability, a deforming interface separates the two layers, with a uniform solute gradient but asymmetric spatially averaged flow intensity. These results highlight the crucial role of diffusivity and viscosity in shaping Marangoni flows and enhance our understanding of interfacial instability dynamics.

physics.flu-dyn↗

arXiv2024

Linear instability in thermally-stratified quasi-Keplerian flows

Quasi-Keplerian flow, a special regime of Taylor-Couette co-rotating flow, is of great astrophysical interest for studying angular momentum transport in accretion disks. The well-known magnetorotational instability (MRI) successfully explains the flow instability and generation of turbulence in certain accretion disks, but fails to account for these phenomena in protoplanetary disks where magnetic effects are negligible. Given the intrinsic decrease of the temperature in these disks, we examine the effect of radial thermal stratification on 3-D global disturbances in linearised quasi-Keplerian flows under radial gravitational acceleration mimicking stellar gravity. Our results show a thermohydrodynamic linear instability for both axisymmetric and non-axisymmetric modes across a broad parameter space of the thermally-stratified quasi-Keplerian flow. Generally, decreasing Richardson or Prandtl numbers stabilises the flow, while a reduced radius ratio destabilises it. This work also provides a quantitative characterisation of the instability. At low Prandtl numbers $Pr$, we observe a scaling relation of the linear critical Taylor number $Ta_c\propto$$Pr^{-6/5}$. Extrapolating the observed scaling to high $Ta$ and low $Pr$ may suggest the relevance of the instability to accretion disks. Moreover, even slight thermal stratification, characterised by a low Richardson number, can trigger the flow instability with a small axial wavelength. These findings are qualitatively consistent with the results from a traditional local stability analysis based on short wave approximations. Our study refines the thermally-induced linearly-unstable transition route in protoplanetary disks to explain angular momentum transport in dead zones where MRI is ineffective.

physics.flu-dyn↗