arXiv subjects

Meng Wu

Publications and source records attributed to Meng Wu.

At least 19 recordsLinked to original sources

arXiv2026

See What I See, Know What I Think: Dense Latent Communication Across Heterogeneous Agents

Multi-agent systems communicate mostly through text, paying a lossy and expensive decode and re-encode cost. KV-cache communication is a promising alternative, yet most prior work is homogeneous, using duplicate copies of the same model, and avoids the central challenge of cross-model latent alignment; existing heterogeneous methods are also restrictive, typically assuming shared input and using transferred caches mainly for steering. We study a more fundamental question: can heterogeneous agents be aligned well enough to perform real "mind reading" and transfer both what one agent sees and how it thinks? Our information-structure analysis reveals a duality: context-aware transfer is driven by sparse reasoning signals, while context-unaware transfer, where the receiver sees no input, requires dense contextual knowledge preservation. Motivated by this, we propose dense alignment for heterogeneous KV-cache communication via a lightweight cross-model cache transformation and two-phase training: reconstruction followed by generation. Across all six directions of {Qwen3-4B, 8B, 14B} and six in-domain and out-of-domain benchmarks, our method outperforms prior heterogeneous baselines, matches or exceeds text communication in context-aware settings at roughly 2 to 3 times lower compute, and remains effective in context-unaware transfer where prior methods collapse.

cs.MA↗

arXiv2026

Design and Analysis Strategies for Pooling in High Throughput Screening: Application to the Search for a New Anti-Microbial

A major public health issue is the growing resistance of bacteria to antibiotics. An important part of the needed response is the discovery and development of new antimicrobial strategies. These require the screening of potential new drugs, typically accomplished using high-throughput screening (HTS). Traditionally, HTS is performed by examining one compound per well, but a more efficient strategy pools multiple compounds per well. In this work, we study several recently proposed pooling construction methods, as well as a variety of pooled high-throughput screening analysis methods, in order to provide guidance to practitioners on which methods to use. This is done in the context of an application of the methods to the search for new drugs to combat bacterial infection. We discuss both an extensive pilot study as well as a small screening campaign, and highlight both the successes and challenges of the pooling approach.

stat.AP↗

arXiv2026

IntPro: A Proxy Agent for Context-Aware Intent Understanding via Retrieval-conditioned Inference

Large language models (LLMs) have become integral to modern Human-AI collaboration workflows, where accurately understanding user intent serves as a crucial step for generating satisfactory responses. Context-aware intent understanding, which involves inferring user intentions from situational environments, is inherently challenging because it requires reasoning over both the immediate context and the user's underlying motivations that drive their behavior. Moreover, existing approaches often treat intent understanding as a static recognition task, overlooking users' accumulated intent patterns that could provide valuable references for more accurate and generalizable understanding. To address this gap, we propose IntPro, a proxy agent that learns to adapt to individual users via retrieval-conditioned intent inference. We design intent explanations that abstract how contextual signals connect to expressed intents, and store them in an individual intent history library for retrieval. We train IntPro through supervised fine-tuning on retrieval-conditioned trajectories and multi-turn Group Relative Policy Optimization (GRPO) with tool-aware reward functions, enabling the agent to learn when to leverage historical intent patterns and when to infer directly. Experiments across three diverse scenarios (Highlight-Intent, MIntRec2.0, and Weibo Post-Sync) demonstrate that IntPro achieves strong intent understanding performance with effective context-aware reasoning capabilities across different scenarios and model types.

cs.CL↗

arXiv2026

SoulSeek: Exploring the Use of Social Cues in LLM-based Information Seeking

Social cues, which convey others' presence, behaviors, or identities, play a crucial role in human information seeking by helping individuals judge relevance and trustworthiness. However, existing LLM-based search systems primarily rely on semantic features, creating a misalignment with the socialized cognition underlying natural information seeking. To address this gap, we explore how the integration of social cues into LLM-based search influences users' perceptions, experiences, and behaviors. Focusing on social media platforms that are beginning to adopt LLM-based search, we integrate design workshops, the implementation of the prototype system (SoulSeek), a between-subjects study, and mixed-method analyses to examine both outcome- and process-level findings. The workshop informs the prototype's cue-integrated design. The study shows that social cues improve perceived outcomes and experiences, promote reflective information behaviors, and reveal limits of current LLM-based search. We propose design implications emphasizing better social-knowledge understanding, personalized cue settings, and controllable interactions.

cs.HC↗

arXiv2025

Generalization of Diffusion Models Arises with a Balanced Representation Space

Diffusion models excel at generating high-quality, diverse samples, yet they risk memorizing training data when overfit to the training objective. We analyze the distinctions between memorization and generalization in diffusion models through the lens of representation learning. By investigating a two-layer ReLU denoising autoencoder (DAE), we prove that (i) memorization corresponds to the model storing raw training samples in the learned weights for encoding and decoding, yielding localized spiky representations, whereas (ii) generalization arises when the model captures local data statistics, producing balanced representations. Furthermore, we validate these theoretical findings on real-world unconditional and text-to-image diffusion models, demonstrating that the same representation structures emerge in deep generative models with significant practical implications. Building on these insights, we propose a representation-based method for detecting memorization and a training-free editing technique that allows precise control via representation steering. Together, our results highlight that learning good representations is central to novel and meaningful generative modeling.

cs.LG↗

arXiv2025

Coarse-to-Fine Hierarchical Alignment for UAV-based Human Detection using Diffusion Models

Training object detectors demands extensive, task-specific annotations, yet this requirement becomes impractical in UAV-based human detection due to constantly shifting target distributions and the scarcity of labeled images. As a remedy, synthetic simulators are adopted to generate annotated data, with a low annotation cost. However, the domain gap between synthetic and real images hinders the model from being effectively applied to the target domain. Accordingly, we introduce Coarse-to-Fine Hierarchical Alignment (CFHA), a three-stage diffusion-based framework designed to transform synthetic data for UAV-based human detection, narrowing the domain gap while preserving the original synthetic labels. CFHA explicitly decouples global style and local content domain discrepancies and bridges those gaps using three modules: (1) Global Style Transfer -- a diffusion model aligns color, illumination, and texture statistics of synthetic images to the realistic style, using only a small real reference set; (2) Local Refinement -- a super-resolution diffusion model is used to facilitate fine-grained and photorealistic details for the small objects, such as human instances, preserving shape and boundary integrity; (3) Hallucination Removal -- a module that filters out human instances whose visual attributes do not align with real-world data to make the human appearance closer to the target distribution. Extensive experiments on public UAV Sim2Real detection benchmarks demonstrate that our methods significantly improve the detection accuracy compared to the non-transformed baselines. Specifically, our method achieves up to $+14.1$ improvement of mAP50 on Semantic-Drone benchmark. Ablation studies confirm the complementary roles of the global and local stages and highlight the importance of hierarchical alignment. The code is released at \href{https://github.com/liwd190019/CFHA}{this url}.

cs.CV↗

arXiv2025

INT v.s. FP: A Comprehensive Study of Fine-Grained Low-bit Quantization Formats

Modern AI hardware, such as Nvidia's Blackwell architecture, is increasingly embracing low-precision floating-point (FP) formats to handle the pervasive activation outliers in Large Language Models (LLMs). Despite this industry trend, a unified comparison of FP and integer (INT) quantization across varying granularities has been missing, leaving algorithm and hardware co-design without clear guidance. This paper fills that gap by systematically investigating the trade-offs between FP and INT formats. We reveal a critical performance crossover: while FP excels in coarse-grained quantization, the comparison at fine-grained (block-wise) levels is more nuanced. Our comprehensive comparison demonstrates that for popular 8-bit fine-grained formats (e.g., MX with block size 32), MXINT8 is superior to its FP counterpart in both algorithmic accuracy and hardware efficiency. However, for 4-bit formats, FP (e.g., MXFP4, NVFP4) often holds an accuracy advantage , though we show that NVINT4 can surpass NVFP4 when outlier-mitigation techniques like Hadamard rotation are applied. We also introduce a symmetric clipping method that resolves gradient bias in fine-grained low-bit INT training, enabling nearly lossless performance for MXINT8 training. These findings challenge the current hardware trajectory, demonstrating that a one-size-fits-all FP approach is suboptimal and advocating that fine-grained INT formats, particularly MXINT8, offer a better balance of accuracy, power, and efficiency for future AI accelerators.

cs.LG↗

arXiv2025

Dynamical control of quantum photon-photon interaction with phase change material

Quantum interference can produce a pivotal effective photon-photon interaction, enabling the exploration of various quantum information technologies that beyond the possibilities of classical physics. While such an effective interaction is fundamentally limited to the bosonic nature of photons and the restricted phase responses from commonly used unitary optical elements, loss-induced nonunitary operation provides an alternative degree of freedom to control the quantum interference. Here, we propose and experimentally demonstrate a concise yet powerful tool to unravel fundamental features of quantum interference based on the phase change material vanadium dioxide. Since the insulator-metal transition in an elaborate vanadium dioxide thin film can create any desired particle exchange phase response, we show its tunability over the effective photon-photon interaction between paired photons that are entangled in the symmetric and anti-symmetric forms, which may introduce sophisticated nonunitary operations and functionalities into programmable optical platforms. These results provide an alternative approach to investigate the quantum light-matter interaction, and facilitate the use of quantum interference for various quantum information processing tasks such as quantum simulation and quantum computation.

quant-ph↗

arXiv2025

A Closer Look at Model Collapse: From a Generalization-to-Memorization Perspective

The widespread use of diffusion models has led to an abundance of AI-generated data, raising concerns about model collapse -- a phenomenon in which recursive iterations of training on synthetic data lead to performance degradation. Prior work primarily characterizes this collapse via variance shrinkage or distribution shift, but these perspectives miss practical manifestations of model collapse. This paper identifies a transition from generalization to memorization during model collapse in diffusion models, where models increasingly replicate training data instead of generating novel content during iterative training on synthetic samples. This transition is directly driven by the declining entropy of the synthetic training data produced in each training cycle, which serves as a clear indicator of model degradation. Motivated by this insight, we propose an entropy-based data selection strategy to mitigate the transition from generalization to memorization and alleviate model collapse. Empirical results show that our approach significantly enhances visual quality and diversity in recursive generation, effectively preventing collapse.

cs.LG↗

arXiv2025

DTVM: Revolutionizing Smart Contract Execution with Determinism and Compatibility

We introduce the DeTerministic Virtual Machine (DTVM) Stack, a next-generation smart contract execution framework designed to address critical performance, determinism, and ecosystem compatibility challenges in blockchain networks. Building upon WebAssembly (Wasm) while maintaining full Ethereum Virtual Machine (EVM) ABI compatibility, DTVM introduces a Deterministic Middle Intermediate Representation (dMIR) and a hybrid lazy-JIT compilation engine to balance compilation speed and execution efficiency. DTVM further accommodates diverse instruction set architectures (e.g., EVM, RISC-V) through modular adaptation layers. This enables seamless integration with DTVM's hybrid lazy-JIT compilation engine, which dynamically optimizes performance while preserving deterministic execution guarantees across heterogeneous environments. The key contributions including: 1). The framework achieves up to 2$\times$ acceleration over evmone in dominant Ethereum contract (e.g. ERC20/721/1155) execution and reduces fibonacci computation latency by 11.8$\sim$40.5% compared to Wasm based VMs. 2). A novel trampoline hot-switch mechanism enables sub-millisecond (0.95ms) post-deployment invocation times, outperforming up to about 23$\times$ in compilation and invocation efficiency. 3). It supports multi-language development (Solidity, C++, Rust, Java, Go, and AssemblyScript) through unified bytecode conversion while maintaining EVM ABI compatibility for seamless invocation. It reduces machine code object sizes by 30.0$\sim$72.6%, coupled with a minimized Trusted Computing Base. 4). It offers SmartCogent, an AI-driven full-stack development experience, leveraging fine-tuned LLMs and retrieval-augmented generation to automate tasks across the smart contract lifecycle: development, debugging, security auditing, and deployment. DTVM Stack has been open-sourced (https://github.com/DTVMStack).

cs.DC↗

arXiv2025

Projection theorems with countably many exceptions and applications to the exact overlaps conjecture

We establish several optimal estimates for exceptional parameters in the projection of fractal measures: (1) For a parametric family of self-similar measures satisfying a transversality condition, the set of parameters leading to a dimension drop is at most countable. (2) For any ergodic CP-distribution $Q$ on $\mathbb{R}^2$, the Hausdorff dimension of its orthogonal projection is $\min\{1, \dim Q\}$ in all but at most countably many directions. Applications of our projection results include: (i) For any planar Borel probability measure with uniform entropy dimension $\alpha$, the packing dimension of its orthogonal projection is at least $\min\{1, \alpha\}$ in all but at most countably many directions. (ii) For any planar set $F$, the Assouad dimension of its orthogonal projection is at least $ \min\{1, \dim_{\rm A} F\} $ in all but at most countably many directions.

math.DS↗

arXiv2025

Leveraging Compute-in-Memory for Efficient Generative Model Inference in TPUs

With the rapid advent of generative models, efficiently deploying these models on specialized hardware has become critical. Tensor Processing Units (TPUs) are designed to accelerate AI workloads, but their high power consumption necessitates innovations for improving efficiency. Compute-in-memory (CIM) has emerged as a promising paradigm with superior area and energy efficiency. In this work, we present a TPU architecture that integrates digital CIM to replace conventional digital systolic arrays in matrix multiply units (MXUs). We first establish a CIM-based TPU architecture model and simulator to evaluate the benefits of CIM for diverse generative model inference. Building upon the observed design insights, we further explore various CIM-based TPU architectural design choices. Up to 44.2% and 33.8% performance improvement for large language model and diffusion transformer inference, and 27.3x reduction in MXU energy consumption can be achieved with different design choices, compared to the baseline TPUv4i architecture.

cs.AR↗

arXiv2024

Variational integrators for stochastic Hamiltonian systems on Lie groups: properties and convergence

We derive variational integrators for stochastic Hamiltonian systems on Lie groups using a discrete version of the stochastic Hamiltonian phase space principle. The structure-preserving properties of the resulting scheme, such as symplecticity, preservation of the Lie-Poisson structure, preservation of the coadjoint orbits, and conservation of Casimir functions, are discussed, along with a discrete Noether theorem for subgroup symmetries. We also consider in detail the case of stochastic Hamiltonian systems with advected quantities, studying the associated structure-preserving properties in relation to semidirect product Lie groups. A full convergence proof for the scheme is provided for the case of the Lie group of rotations. Several numerical examples are presented, including simulations of the free rigid body and the heavy top.

math.NA↗

arXiv2024

Multilayer Dataflow: Orchestrate Butterfly Sparsity to Accelerate Attention Computation

Recent neural networks (NNs) with self-attention exhibit competitiveness across different AI domains, but the essential attention mechanism brings massive computation and memory demands. To this end, various sparsity patterns are introduced to reduce the quadratic computation complexity, among which the structured butterfly sparsity has been proven efficient in computation reduction while maintaining model accuracy. However, its complicated data accessing pattern brings utilization degradation and makes parallelism hard to exploit in general block-oriented architecture like GPU. Since the reconfigurable dataflow architecture is known to have better data reusability and architectural flexibility in general NN-based acceleration, we want to apply it to the butterfly sparsity for acquiring better computational efficiency for attention workloads. We first propose a hybrid butterfly-sparsity network to obtain better trade-offs between attention accuracy and performance. Next, we propose a scalable multilayer dataflow method supported by coarse-grained streaming parallelism designs, to orchestrate the butterfly sparsity computation on the dataflow array. The experiments show that compared with Jetson Xavier NX, our design has a speedup of up to $14.34\times$ ($9.29\times$ on average) as well as $11.14\times$ energy efficiency advancement in attention workloads. In comparison with SOTA attention accelerators of the same peak performance, our dataflow architecture acquires $2.38\times$-$4.7\times$ efficiency improvement as well as $6.60\times$-$15.37\times$ energy reduction with butterfly sparsity optimization.

cs.AR↗

arXiv2024

New results on embeddings of self-similar sets via renormalization

For self-similar sets $X,Y\subseteq \mathbb{R}$, we obtain new results towards the affine embeddings conjecture of Feng-Huang-Rao (2014), and the equivalent weak intersections conjecture. We show that the conjecture holds when the defining maps of $X,Y$ have algebraic contraction ratios, and also for arbitrary $Y$ when the maps defining $X$ have algebraic-contraction ratios and there are sufficiently many of them relative to the number of maps defining $Y$. We also show that it holds for a.e. choice of the corresponding contraction ratios, and obtain bounds on the packing dimension of the exceptional parameters. Key ingredients in our argument include a new renormalization procedure in the space of embeddings, and Orponen's projection Theorem for Assouad dimension (2021).

math.DS↗

arXiv2024

UniCoN: Universal Conditional Networks for Multi-Age Embryonic Cartilage Segmentation with Sparsely Annotated Data

Osteochondrodysplasia, affecting 2-3% of newborns globally, is a group of bone and cartilage disorders that often result in head malformations, contributing to childhood morbidity and reduced quality of life. Current research on this disease using mouse models faces challenges since it involves accurately segmenting the developing cartilage in 3D micro-CT images of embryonic mice. Tackling this segmentation task with deep learning (DL) methods is laborious due to the big burden of manual image annotation, expensive due to the high acquisition costs of 3D micro-CT images, and difficult due to embryonic cartilage's complex and rapidly changing shapes. While DL approaches have been proposed to automate cartilage segmentation, most such models have limited accuracy and generalizability, especially across data from different embryonic age groups. To address these limitations, we propose novel DL methods that can be adopted by any DL architectures -- including CNNs, Transformers, or hybrid models -- which effectively leverage age and spatial information to enhance model performance. Specifically, we propose two new mechanisms, one conditioned on discrete age categories and the other on continuous image crop locations, to enable an accurate representation of cartilage shape changes across ages and local shape details throughout the cranial region. Extensive experiments on multi-age cartilage segmentation datasets show significant and consistent performance improvements when integrating our conditional modules into popular DL segmentation architectures. On average, we achieve a 1.7% Dice score increase with minimal computational overhead and a 7.5% improvement on unseen data. These results highlight the potential of our approach for developing robust, universal models capable of handling diverse datasets with limited annotated data, a key challenge in DL-based medical image analysis.

eess.IV↗

arXiv2024

Geometric, Variational, and Bracket Descriptions of Fluid Motion with Open Boundaries

We develop a Lie group geometric framework for the motion of fluids with permeable boundaries that extends Arnold's geometric description of fluid in closed domains. Our setting is based on the classical Hamilton principle applied to fluid trajectories, appropriately amended to incorporate bulk and boundary forces via a Lagrange-d'Alembert approach, and to take into account only the fluid particles present in the fluid domain at a given time. By applying a reduction by relabelling symmetries, we deduce a variational formulation in the Eulerian description that extends the Euler-Poincar\'e framework to open fluids. On the Hamiltonian side, our approach yields a bracket formulation that consistently extends the Lie-Poisson bracket of fluid dynamics and contains as a particular case a bulk+boundary bracket formulation proposed earlier. We illustrate the geometric framework with several examples, and also consider the extension of this setting to include multicomponent fluids, the consideration of general advected quantities, the analysis of higher-order fluids, and the incorporation of boundary stresses. Open fluids are found in a wide range of physical systems, including geophysical fluid dynamics, porous media, incompressible flow and magnetohydrodynamics. This new formulation permits a description based on geometric mechanics that can be applied to a broad class of models.

math.DS↗

arXiv2024

Accelerating Mini-batch HGNN Training by Reducing CUDA Kernels

Heterogeneous graph neural networks (HGNNs) are essential for capturing the structure and semantic information in heterogeneous graphs. However, existing GPU-based solutions, such as PyTorch Geometric, suffer from low GPU utilization due to numerous short-execution-time and memory-bound CUDA kernels during HGNN training. To address this issue, we introduce HiFuse, an enhancement for PyTorch Geometric designed to accelerate mini-batch HGNN training on CPU-GPU systems. From the data perspective, we reorganize and merge multiple smaller vertex feature matrices into larger ones, enabling a single kernel to process larger data chunks. This efficiently exploits data locality, reduces the kernel launch overhead, and improves overall GPU utilization. From the workflow perspective, we sophisticatedly offload the construction of semantic graphs from GPU to CPU to reduce the number of CUDA kernels. To meet the parallelism requirements on CPU and ensure seamless execution between CPU and GPU, we employ parallelization techniques including multi-threading and asynchronous pipeline. This allows different stages of the process to overlap, enhancing GPU utilization and reducing end-to-end execution latency, leading to a more efficient and balanced use of computational resources. Through extensive experiments, HiFuse demonstrates an average 2.38 times speedup compared to a state-of-the-art solution.

cs.AR↗