arXiv subjects

Yifan Zhu

Publications and source records attributed to Yifan Zhu.

At least 73 records · Page 4Linked to original sources

arXiv2026

Single-pulse Stimulated Raman Photothermal Microscopy and Direct Visualization of Cholesterol-rich Membrane Domains

By measuring the thermal effects resulting from stimulated Raman excitations, stimulated Raman photothermal (SRP) microscope offers a new access to spatially and temporally resolved Raman signatures across various sample types. Unlike stimulated Raman scattering (SRS) microscopy, SRP is highly compatible with noisy ultrafast laser sources, allowing the use of high peak power, low repetition rate optical parametric amplifier (OPA) lasers to boost sensitivity and imaging speed. Here, we report a single pulse SRP (spSRP) microscope system in which SRP signals are induced by individual pairs of laser pulses generated by an OPA laser. Extensive pulse chirping is used to maximize SRS excitation rate and to minimize photodamage. The single-pixel limit of detection (LOD) of spSRP on dimethyl sulfoxide (DMSO) reaches 890 μM, which is a ~44-fold improvement over SRS microscope. Versatile applications to fungi, cells, and tissues are demonstrated. Live cell spSRP imaging was carried out at speed of 10 frames per second. Particularly, the spSRP system enables direct visualization of cholesterol-rich domains, co-localized with caveolin immunofluorescence, in the plasma membrane of HeLa cells.

physics.optics↗

arXiv2026

Micromolar chemical imaging by high-energy low-photodamage Coherent Anti-stokes Raman Scattering (HELP-CARS)

Coherent anti-Stokes Raman scattering (CARS) microscopy offers label-free chemical imaging capabilities, but its performance is constrained by small Raman scattering cross-section, strong non-resonant background (NRB), and limited signal-to-noise ratio (SNR). Here, we introduce a high-energy, low-photodamage CARS (HELP-CARS) platform designed to overcome these physical limitations. By employing a 1-MHz non-collinear optical parametric amplifier (NOPA) with extensive pulse chirping, HELP-CARS increases the coherent Raman excitation efficiency by ~300 times and improves the signal-to-nonresonant background ratio by 11 times, while inducing negligible damage during live cell imaging. Furthermore, to remove non-independent noise and physically entangled non-resonant background, we incorporate self-supervised deep-learning denoising and background removal based on the Kramers-Kronig relationship, yielding sensitivity improvement by an additional order of magnitude. Together, these advances provide a micromolar imaging sensitivity (160 uM for Dimethyl sulfoxide-d6) corresponding to 1000 molecules in the focal volume. Such high sensitivity enables high-fidelity chemical imaging in both fingerprint and silent windows. Hyperspectral HELP-CARS imaging of deuterated fatty acids allowed first observation of chemical separation with single lipid droplet. Together, HELP-CARS offers a powerful and generalizable approach for ultrasensitive and quantitative vibrational imaging of biological systems.

physics.optics↗

arXiv2026

Token-Guard: Towards Token-Level Hallucination Control via Self-Checking Decoding

Large Language Models (LLMs) often hallucinate, generating content inconsistent with the input. Retrieval-Augmented Generation (RAG) and Reinforcement Learning with Human Feedback (RLHF) can mitigate hallucinations but require resource-intensive retrieval or large-scale fine-tuning. Decoding-based methods are lighter yet lack explicit hallucination control. To address this, we present Token-Guard, a token-level hallucination control method based on self-checking decoding. Token-Guard performs internal verification at each reasoning step to detect hallucinated tokens before they propagate. Candidate fragments are further evaluated in a latent space with explicit hallucination risk scoring, while iterative pruning and regeneration dynamically correct detected errors. Experiments on HALU datasets show Token-Guard substantially reduces hallucinations and improves generation accuracy, offering a scalable, modular solution for reliable LLM outputs. Our code is publicly available.

cs.CL↗

arXiv2026

Optimisation of the vertex detector and measurement of Higgs decays to second-generation quarks at the CEPC

The vertex detector is crucial for precision measurements of the Higgs boson at the electron-positron Higgs factory. Benchmarked with $H \to c\bar{c}$ and $H \to s\bar{s}$ measurements in the $ν\barνH$ channel, we perform an optimisation study on the inner radius and spatial resolution of the vertex detector using the Jet Origin Identification (JOI) framework, which determines the parton flavor of jets using advanced Artificial Intelligence (AI) algorithm. We observe that, compared to the reference detector configuration, halving the inner radius and spatial resolution improves the transverse and longitudinal impact parameter resolution approximately by a factor of two, while increasing the accuracy and significance of the $H \to c\bar{c}/s\bar{s}$ measurement by 4\% and 8\%, respectively. Conversely, doubling these parameters results in comparable degradation, with variations in the inner radius being the dominant factor. Our results provide guidance for detector design and highlight promising prospects for identifying the $H \to s\bar{s}$ decay mode at future Higgs factories.

hep-ex↗

arXiv2026

Sawtooth Wavefront Reordering: Enhanced CuTile FlashAttention on NVIDIA GB10

High-performance attention kernels are essential for Large Language Models. This paper presents analysis of CuTile-based Flash Attention memory behavior and a technique to improve its cache performance. In particular, our analysis on the NVIDIA GB10 (Grace Blackwell) identifies the main cause of L2 cache miss. Leveraging this insight, we introduce a new programming technique called Sawtooth Wavefront Reordering that reduces L2 misses. We validate it in both CUDA and CuTile, observing 50\% or greater reduction in L2 misses and up to 60\% increase in throughput on GB10.

cs.PF↗

arXiv2026

Vision Language Models for Optimization-Driven Intent Processing in Autonomous Networks

Intent-Based Networking (IBN) allows operators to specify high-level network goals rather than low-level configurations. While recent work demonstrates that large language models can automate configuration tasks, a distinct class of intents requires generating optimization code to compute provably optimal solutions for traffic engineering, routing, and resource allocation. Current systems assume text-based intent expression, requiring operators to enumerate topologies and parameters in prose. Network practitioners naturally reason about structure through diagrams, yet whether Vision-Language Models (VLMs) can process annotated network sketches into correct optimization code remains unexplored. We present IntentOpt, a benchmark of 85 optimization problems across 17 categories, evaluating four VLMs (GPT-5-Mini, Claude-Haiku-4.5, Gemini-2.5-Flash, Llama-3.2-11B-Vision) under three prompting strategies on multimodal versus text-only inputs. Our evaluation shows that visual parameter extraction reduces execution success by 12-21 percentage points (pp), with GPT-5-Mini dropping from 93% to 72%. Program-of-thought prompting decreases performance by up to 13 pp, and open-source models lag behind closed-source ones, with Llama-3.2-11B-Vision reaching 18% compared to 75% for GPT-5-Mini. These results establish baseline capabilities and limitations of current VLMs for optimization code generation within an IBN system. We also demonstrate practical feasibility through a case study that deploys VLM-generated code to network testbed infrastructure using Model Context Protocol.

cs.AI↗

arXiv2026

DecoupledESC: Enhancing Emotional Support Generation via Strategy-Response Decoupled Preference Optimization

Recent advances in Emotional Support Conversation (ESC) have improved emotional support generation by fine-tuning Large Language Models (LLMs) via Supervised Fine-Tuning (SFT). However, common psychological errors still persist. While Direct Preference Optimization (DPO) shows promise in reducing such errors through pairwise preference learning, its effectiveness in ESC tasks is limited by two key challenges: (1) Entangled data structure: Existing ESC data inherently entangles psychological strategies and response content, making it difficult to construct high-quality preference pairs; and (2) Optimization ambiguity: Applying vanilla DPO to such entangled pairwise data leads to ambiguous training objectives. To address these issues, we introduce Inferential Preference Mining (IPM) to construct high-quality preference data, forming the IPM-PrefDial dataset. Building upon this data, we propose a Decoupled ESC framework inspired by Gross's Extended Process Model of Emotion Regulation, which decomposes the ESC task into two sequential subtasks: strategy planning and empathic response generation. Each was trained via SFT and subsequently enhanced by DPO to align with the psychological preference. Extensive experiments demonstrate that our Decoupled ESC framework outperforms joint optimization baselines, reducing preference bias and improving response quality.

cs.CL↗

arXiv2026

Branch, or Layer? Zeroth-Order Optimization for Continual Learning of Vision-Language Models

Vision-Language Continual Learning (VLCL) has attracted significant research attention for its robust capabilities, and the adoption of Parameter-Efficient Fine-Tuning (PEFT) strategies is enabling these models to achieve competitive performance with substantially reduced resource consumption. However, dominated First-Order (FO) optimization is prone to trap models in suboptimal local minima, especially in limited exploration subspace within PEFT. To overcome this challenge, this paper pioneers a systematic exploration of adopting Zeroth-Order (ZO) optimization for PEFT-based VLCL. We first identify the incompatibility of naive full-ZO adoption in VLCL due to optimization process instability. We then investigate the application of ZO optimization from a modality branch-wise to a fine-grained layer-wise across various training units to identify an optimal strategy. Besides, a key theoretical insight reveals that vision modality exhibit higher variance than language counterparts in VLCL during the ZO optimization process, and we propose a modality-aware ZO strategy, which adopts gradient sign normalization in ZO and constrains vision modality perturbation to further improve performance. Benefiting from the adoption of ZO optimization, PEFT-based VLCL fulfills better ability to escape local minima during the optimization process, extensive experiments on four benchmarks demonstrate that our method achieves state-of-the-art results.

cs.CV↗

arXiv2026

All-in-one Graph-based Indexing for Hybrid Search on GPUs

Hybrid search has emerged as a promising paradigm that combines lexical and semantic retrieval, enhancing accuracy for applications such as recommendations, information retrieval, and Retrieval-Augmented Generation. However, existing methods are constrained by a trilemma: they sacrifice flexibility for efficiency, suffer from accuracy degradation, or incur prohibitive storage overhead for flexible combinations of retrieval paths. This paper introduces Allan-Poe, a novel all-in-one graph index accelerated by GPUs for efficient hybrid search. We first analyze the limitations of existing retrieval paradigms and extract key design principles for an effective hybrid index. Guided by the principles, we architect a unified graph-based index that flexibly integrates three retrieval paths (dense vector, sparse vector, and full-text) within a single, cohesive structure. To enable efficient construction, we design a GPU-accelerated pipeline featuring a warp-level hybrid distance kernel, RNG-IP joint pruning, and keyword-aware neighbor recycling. For query processing, we introduce a dynamic fusion framework that supports any combination of retrieval paths and weights without index reconstruction, flexibly leveraging logical structures from the knowledge graph to resolve complex multi-hop queries. Extensive experiments on 6 real-world datasets demonstrate that Allan-Poe achieves superior end-to-end query accuracy and outperforms state-of-the-art methods by 1.5x-186.4x in throughput, while significantly reducing storage overhead.

cs.DB↗

arXiv2025

UniCom: Towards a Unified and Cohesiveness-aware Framework for Community Search and Detection

Searching and detecting communities in real-world graphs underpins a wide range of applications. Despite the success achieved, current learning-based solutions regard community search, i.e., locating the best community for a given query, and community detection, i.e., partitioning the whole graph, as separate problems, necessitating task- and dataset-specific retraining. Such a strategy limits the applicability and generalization ability of the existing models. Additionally, these methods rely heavily on information from the target dataset, leading to suboptimal performance when supervision is limited or unavailable. To mitigate this limitation, we propose UniCom, a unified framework to solve both community search and detection tasks through knowledge transfer across multiple domains, thus alleviating the limitations of single-dataset learning. UniCom centers on a Domain-aware Specialization (DAS) procedure that adapts on the fly to unseen graphs or tasks, eliminating costly retraining while maintaining framework compactness with a lightweight prompt-based paradigm. This is empowered by a Universal Graph Learning (UGL) backbone, which distills transferable semantic and topological knowledge from multiple source domains via comprehensive pre-training. Both DAS and UGL are informed by local neighborhood signals and cohesive subgraph structures, providing consistent guidance throughout the framework. Extensive experiments on both tasks across 16 benchmark datasets and 22 baselines have been conducted to ensure a comprehensive and fair evaluation. UniCom consistently outperforms all state-of-the-art baselines across all tasks under settings with scarce or no supervision, while maintaining runtime efficiency.

cs.SI↗

arXiv2025

CreBench: Human-Aligned Creativity Evaluation from Idea to Process to Product

Human-defined creativity is highly abstract, posing a challenge for multimodal large language models (MLLMs) to comprehend and assess creativity that aligns with human judgments. The absence of an existing benchmark further exacerbates this dilemma. To this end, we propose CreBench, which consists of two key components: 1) an evaluation benchmark covering the multiple dimensions from creative idea to process to products; 2) CreMIT (Creativity Multimodal Instruction Tuning dataset), a multimodal creativity evaluation dataset, consisting of 2.2K diverse-sourced multimodal data, 79.2K human feedbacks and 4.7M multi-typed instructions. Specifically, to ensure MLLMs can handle diverse creativity-related queries, we prompt GPT to refine these human feedbacks to activate stronger creativity assessment capabilities. CreBench serves as a foundation for building MLLMs that understand human-aligned creativity. Based on the CreBench, we fine-tune open-source general MLLMs, resulting in CreExpert, a multimodal creativity evaluation expert model. Extensive experiments demonstrate that the proposed CreExpert models achieve significantly better alignment with human creativity evaluation compared to state-of-the-art MLLMs, including the most advanced GPT-4V and Gemini-Pro-Vision.

cs.AI↗

arXiv2025

More Than Irrational: Modeling Belief-Biased Agents

Despite the explosive growth of AI and the technologies built upon it, predicting and inferring the sub-optimal behavior of users or human collaborators remains a critical challenge. In many cases, such behaviors are not a result of irrationality, but rather a rational decision made given inherent cognitive bounds and biased beliefs about the world. In this paper, we formally introduce a class of computational-rational (CR) user models for cognitively-bounded agents acting optimally under biased beliefs. The key novelty lies in explicitly modeling how a bounded memory process leads to a dynamically inconsistent and biased belief state and, consequently, sub-optimal sequential decision-making. We address the challenge of identifying the latent user-specific bound and inferring biased belief states from passive observations on the fly. We argue that for our formalized CR model family with an explicit and parameterized cognitive process, this challenge is tractable. To support our claim, we propose an efficient online inference method based on nested particle filtering that simultaneously tracks the user's latent belief state and estimates the unknown cognitive bound from a stream of observed actions. We validate our approach in a representative navigation task using memory decay as an example of a cognitive bound. With simulations, we show that (1) our CR model generates intuitively plausible behaviors corresponding to different levels of memory capacity, and (2) our inference method accurately and efficiently recovers the ground-truth cognitive bounds from limited observations ($\le 100$ steps). We further demonstrate how this approach provides a principled foundation for developing adaptive AI assistants, enabling adaptive assistance that accounts for the user's memory limitations.

cs.AI↗

arXiv2025

Balancing the Blend: An Experimental Analysis of Trade-offs in Hybrid Search

Hybrid search, the integration of lexical and semantic retrieval, has become a cornerstone of modern information retrieval systems, driven by demanding applications like Retrieval-Augmented Generation (RAG). The architectural design space for these systems is vast and complex, yet a systematic understanding of the trade-offs among their core components -- retrieval paradigms, combination schemes, and re-ranking methods -- is lacking. To address this, and informed by our experience building the Infinity open-source database, we present the first experimental analysis of advanced hybrid search architectures. Our framework integrates four retrieval paradigms -- Full-Text Search (FTS), Sparse Vector Search (SVS), Dense Vector Search (DVS), and Tensor Search (TenS) -- and evaluates their combinations and re-ranking strategies across 11 real-world datasets. Our results reveal three key findings: (1) A "weakest link" phenomenon, where a weak path can substantially degrade overall accuracy, highlighting the need for path-wise quality assessment before fusion. (2) A data-driven map of performance trade-offs, demonstrating that optimal configurations depend heavily on resource constraints and data characteristics, precluding a one-size-fits-all solution. (3) The identification of Tensor-based Re-ranking Fusion (TRF) as a high-efficacy alternative to mainstream fusion methods, offering the semantic power of tensor search at a fraction of the computational and memory cost. Our findings offer concrete guidelines for designing adaptive, scalable hybrid search systems and identify key directions for future research.

cs.DB↗

arXiv2025

Seed3D 1.0: From Images to High-Fidelity Simulation-Ready 3D Assets

Developing embodied AI agents requires scalable training environments that balance content diversity with physics accuracy. World simulators provide such environments but face distinct limitations: video-based methods generate diverse content but lack real-time physics feedback for interactive learning, while physics-based engines provide accurate dynamics but face scalability limitations from costly manual asset creation. We present Seed3D 1.0, a foundation model that generates simulation-ready 3D assets from single images, addressing the scalability challenge while maintaining physics rigor. Unlike existing 3D generation models, our system produces assets with accurate geometry, well-aligned textures, and realistic physically-based materials. These assets can be directly integrated into physics engines with minimal configuration, enabling deployment in robotic manipulation and simulation training. Beyond individual objects, the system scales to complete scene generation through assembling objects into coherent environments. By enabling scalable simulation-ready content creation, Seed3D 1.0 provides a foundation for advancing physics-based world simulators. Seed3D 1.0 is now available on https://console.volcengine.com/ark/region:ark+cn-beijing/experience/vision?modelId=doubao-seed3d-1-0-250928&tab=Gen3D

eess.IV↗

arXiv2025

HyperGraphRAG: Retrieval-Augmented Generation via Hypergraph-Structured Knowledge Representation

Standard Retrieval-Augmented Generation (RAG) relies on chunk-based retrieval, whereas GraphRAG advances this approach by graph-based knowledge representation. However, existing graph-based RAG approaches are constrained by binary relations, as each edge in an ordinary graph connects only two entities, limiting their ability to represent the n-ary relations (n >= 2) in real-world knowledge. In this work, we propose HyperGraphRAG, a novel hypergraph-based RAG method that represents n-ary relational facts via hyperedges, and consists of knowledge hypergraph construction, retrieval, and generation. Experiments across medicine, agriculture, computer science, and law demonstrate that HyperGraphRAG outperforms both standard RAG and previous graph-based RAG methods in answer accuracy, retrieval efficiency, and generation quality. Our data and code are publicly available at https://github.com/LHRLAB/HyperGraphRAG.

cs.AI↗

arXiv2025

Provable Watermarking for Data Poisoning Attacks

In recent years, data poisoning attacks have been increasingly designed to appear harmless and even beneficial, often with the intention of verifying dataset ownership or safeguarding private data from unauthorized use. However, these developments have the potential to cause misunderstandings and conflicts, as data poisoning has traditionally been regarded as a security threat to machine learning systems. To address this issue, it is imperative for harmless poisoning generators to claim ownership of their generated datasets, enabling users to identify potential poisoning to prevent misuse. In this paper, we propose the deployment of watermarking schemes as a solution to this challenge. We introduce two provable and practical watermarking approaches for data poisoning: {\em post-poisoning watermarking} and {\em poisoning-concurrent watermarking}. Our analyses demonstrate that when the watermarking length is $Θ(\sqrt{d}/ε_w)$ for post-poisoning watermarking, and falls within the range of $Θ(1/ε_w^2)$ to $O(\sqrt{d}/ε_p)$ for poisoning-concurrent watermarking, the watermarked poisoning dataset provably ensures both watermarking detectability and poisoning utility, certifying the practicality of watermarking under data poisoning attacks. We validate our theoretical findings through experiments on several attacks, models, and datasets.

cs.CR↗

arXiv2025

Multi-Channel Amplitude-Phase Asymmetric-Encrypted Janus Acoustic Meta-Holograms

Encrypted optical and acoustic meta-holograms only focus on the encrypted hologram in a single channel, viz. modulating spatial amplitude to project a holographic image. In this research, the unique concept of multi-channel amplitude-phase asymmetric-encrypted Janus acoustic meta-holograms is proposed, demonstrating remarkable capabilities of generating, encrypting, and decrypting both amplitude and phase holographic images on both sides of a metascreen. The flexible and decoupled manipulation mechanism for the amplitude-phase of the bidirectional acoustic waves used in our concept offers multiple possibilities to apply various encryption methods. In this work, our system enables single-input, two-faced four-channel asymmetric encryption, which substantially increase the communication capacity of conventional acoustic holograms, and establish a security framework based on mathematical problem, proving its security. Our work can lead to concrete applications including, but not limited to, multi-channel acoustic field communications and acoustic illusion and cloaking in non-transparent media.

physics.app-ph↗

arXiv2025

SCI-Reason: A Dataset with Chain-of-Thought Rationales for Complex Multimodal Reasoning in Academic Areas

Large Language Models (LLMs) and Large Multimodal Models (LMMs) demonstrate impressive problem-solving skills in many tasks and domains. However, their ability to reason with complex images in academic domains has not been systematically investigated. To bridge this gap, we present SCI-Reason, a dataset for complex multimodel reasoning in academic areas. SCI-Reason aims to test and improve the reasoning ability of large multimodal models using real complex images in academic domains. The dataset contains 12,066 images and 12,626 question-answer pairs extracted from PubMed, divided into training, validation and test splits. Each question-answer pair also contains an accurate and efficient inference chain as a guide to improving the inference properties of the dataset. With SCI-Reason, we performed a comprehensive evaluation of 8 well-known models. The best performing model, Claude-3.7-Sonnet, only achieved an accuracy of 55.19%. Error analysis shows that more than half of the model failures are due to breakdowns in multi-step inference chains rather than errors in primary visual feature extraction. This finding underscores the inherent limitations in reasoning capabilities exhibited by current multimodal models when processing complex image analysis tasks within authentic academic contexts. Experiments on open-source models show that SCI-Reason not only enhances reasoning ability but also demonstrates cross-domain generalization in VQA tasks. We also explore future applications of model inference capabilities in this domain, highlighting its potential for future research.

cs.MM↗