arXiv subjects

Yu Sun

Publications and source records attributed to Yu Sun.

At least 145 records · Page 8Linked to original sources

arXiv2025

Super-Resolution Optical Coherence Tomography Using Diffusion Model-Based Plug-and-Play Priors

We propose an OCT super-resolution framework based on a plug-and-play diffusion model (PnP-DM) to reconstruct high-quality images from sparse measurements (OCT B-mode corneal images). Our method formulates reconstruction as an inverse problem, combining a diffusion prior with Markov chain Monte Carlo sampling for efficient posterior inference. We collect high-speed under-sampled B-mode corneal images and apply a deep learning-based up-sampling pipeline to build realistic training pairs. Evaluations on in vivo and ex vivo fish-eye corneal models show that PnP-DM outperforms conventional 2D-UNet baselines, producing sharper structures and better noise suppression. This approach advances high-fidelity OCT imaging in high-speed acquisition for clinical applications.

eess.IV↗

arXiv2025

Ultrasound Report Generation with Multimodal Large Language Models for Standardized Texts

Ultrasound (US) report generation is a challenging task due to the variability of US images, operator dependence, and the need for standardized text. Unlike X-ray and CT, US imaging lacks consistent datasets, making automation difficult. In this study, we propose a unified framework for multi-organ and multilingual US report generation, integrating fragment-based multilingual training and leveraging the standardized nature of US reports. By aligning modular text fragments with diverse imaging data and curating a bilingual English-Chinese dataset, the method achieves consistent and clinically accurate text generation across organ sites and languages. Fine-tuning with selective unfreezing of the vision transformer (ViT) further improves text-image alignment. Compared to the previous state-of-the-art KMVE method, our approach achieves relative gains of about 2\% in BLEU scores, approximately 3\% in ROUGE-L, and about 15\% in CIDEr, while significantly reducing errors such as missing or incorrect content. By unifying multi-organ and multi-language report generation into a single, scalable framework, this work demonstrates strong potential for real-world clinical workflows.

eess.IV↗

arXiv2025

Proxy Prompt: Endowing SAM and SAM 2 with Auto-Interactive-Prompt for Medical Segmentation

In this paper, we aim to address the unmet demand for automated prompting and enhanced human-model interactions of SAM and SAM2 for the sake of promoting their widespread clinical adoption. Specifically, we propose Proxy Prompt (PP), auto-generated by leveraging non-target data with a pre-annotated mask. We devise a novel 3-step context-selection strategy for adaptively selecting the most representative contextual information from non-target data via vision mamba and selective maps, empowering the guiding capability of non-target image-mask pairs for segmentation on target image/video data. To reinforce human-model interactions in PP, we further propose a contextual colorization module via a dual-reverse cross-attention to enhance interactions between target features and contextual-embedding with amplifying distinctive features of user-defined object(s). Via extensive evaluations, our method achieves state-of-the-art performance on four public datasets and yields comparable results with fully-trained models, even when trained with only 16 image masks.

eess.IV↗

arXiv2025

Large AI Model for Delay-Doppler Domain Channel Prediction in 6G OTFS-Based Vehicular Networks

Channel prediction is crucial for high-mobility vehicular networks, as it enables the anticipation of future channel conditions and the proactive adjustment of communication strategies. However, achieving accurate vehicular channel prediction is challenging due to significant Doppler effects and rapid channel variations resulting from high-speed vehicle movement and complex propagation environments. In this paper, we propose a novel delay-Doppler (DD) domain channel prediction framework tailored for high-mobility vehicular networks. By transforming the channel representation into the DD domain, we obtain an intuitive, sparse, and stable depiction that closely aligns with the underlying physical propagation processes, effectively reducing the complex vehicular channel to a set of time-series parameters with enhanced predictability. Furthermore, we leverage the large artificial intelligence (AI) model to predict these DD-domain time-series parameters, capitalizing on their advanced ability to model temporal correlations. The zero-shot capability of the pre-trained large AI model facilitates accurate channel predictions without requiring task-specific training, while subsequent fine-tuning on specific vehicular channel data further improves prediction accuracy. Extensive simulation results demonstrate the effectiveness of our DD-domain channel prediction framework and the superior accuracy of the large AI model in predicting time-series channel parameters, thereby highlighting the potential of our approach for robust vehicular communication systems.

eess.SP↗

arXiv2025

An LLM-Driven Multi-Agent Debate System for Mendelian Diseases

Accurate diagnosis of Mendelian diseases is crucial for precision therapy and assistance in preimplantation genetic diagnosis. However, existing methods often fall short of clinical standards or depend on extensive datasets to build pretrained machine learning models. To address this, we introduce an innovative LLM-Driven multi-agent debate system (MD2GPS) with natural language explanations of the diagnostic results. It utilizes a language model to transform results from data-driven and knowledge-driven agents into natural language, then fostering a debate between these two specialized agents. This system has been tested on 1,185 samples across four independent datasets, enhancing the TOP1 accuracy from 42.9% to 66% on average. Additionally, in a challenging cohort of 72 cases, MD2GPS identified potential pathogenic genes in 12 patients, reducing the diagnostic time by 90%. The methods within each module of this multi-agent debate system are also replaceable, facilitating its adaptation for diagnosing and researching other complex diseases.

q-bio.GN↗

arXiv2025

Developing Modular Grasping and Manipulation Pipeline Infrastructure to Streamline Performance Benchmarking

The robot manipulation ecosystem currently faces issues with integrating open-source components and reproducing results. This limits the ability of the community to benchmark and compare the performance of different solutions to one another in an effective manner, instead relying on largely holistic evaluations. As part of the COMPARE Ecosystem project, we are developing modular grasping and manipulation pipeline infrastructure in order to streamline performance benchmarking. The infrastructure will be used towards the establishment of standards and guidelines for modularity and improved open-source development and benchmarking. This paper provides a high-level overview of the architecture of the pipeline infrastructure, experiments conducted to exercise it during development, and future work to expand its modularity.

cs.RO↗

arXiv2025

One-Minute Video Generation with Test-Time Training

Transformers today still struggle to generate one-minute videos because self-attention layers are inefficient for long context. Alternatives such as Mamba layers struggle with complex multi-scene stories because their hidden states are less expressive. We experiment with Test-Time Training (TTT) layers, whose hidden states themselves can be neural networks, therefore more expressive. Adding TTT layers into a pre-trained Transformer enables it to generate one-minute videos from text storyboards. For proof of concept, we curate a dataset based on Tom and Jerry cartoons. Compared to baselines such as Mamba~2, Gated DeltaNet, and sliding-window attention layers, TTT layers generate much more coherent videos that tell complex stories, leading by 34 Elo points in a human evaluation of 100 videos per method. Although promising, results still contain artifacts, likely due to the limited capability of the pre-trained 5B model. The efficiency of our implementation can also be improved. We have only experimented with one-minute videos due to resource constraints, but the approach can be extended to longer videos and more complex stories. Sample videos, code and annotations are available at: https://test-time-training.github.io/video-dit

cs.CV↗

arXiv2025

Benchmarking Multi-Object Grasping

In this work, we describe a multi-object grasping benchmark to evaluate the grasping and manipulation capabilities of robotic systems in both pile and surface scenarios. The benchmark introduces three robot multi-object grasping benchmarking protocols designed to challenge different aspects of robotic manipulation. These protocols are: 1) the Only-Pick-Once protocol, which assesses the robot's ability to efficiently pick multiple objects in a single attempt; 2) the Accurate pick-trnsferring protocol, which evaluates the robot's capacity to selectively grasp and transport a specific number of objects from a cluttered environment; and 3) the Pick-transferring-all protocol, which challenges the robot to clear an entire scene by sequentially grasping and transferring all available objects. These protocols are intended to be adopted by the broader robotics research community, providing a standardized method to assess and compare robotic systems' performance in multi-object grasping tasks. We establish baselines for these protocols using standard planning and perception algorithms on a Barrett hand, Robotiq parallel jar gripper, and the Pisa/IIT Softhand-2, which is a soft underactuated robotic hand. We discuss the results in relation to human performance in similar tasks we well.

cs.RO↗

arXiv2025

STAR: A Foundation Model-driven Framework for Robust Task Planning and Failure Recovery in Robotic Systems

Modern robotic systems, deployed across domains from industrial automation to domestic assistance, face a critical challenge: executing tasks with precision and adaptability in dynamic, unpredictable environments. To address this, we propose STAR (Smart Task Adaptation and Recovery), a novel framework that synergizes Foundation Models (FMs) with dynamically expanding Knowledge Graphs (KGs) to enable resilient task planning and autonomous failure recovery. While FMs offer remarkable generalization and contextual reasoning, their limitations, including computational inefficiency, hallucinations, and output inconsistencies hinder reliable deployment. STAR mitigates these issues by embedding learned knowledge into structured, reusable KGs, which streamline information retrieval, reduce redundant FM computations, and provide precise, scenario-specific insights. The framework leverages FM-driven reasoning to diagnose failures, generate context-aware recovery strategies, and execute corrective actions without human intervention or system restarts. Unlike conventional approaches that rely on rigid protocols, STAR dynamically expands its KG with experiential knowledge, ensuring continuous adaptation to novel scenarios. To evaluate the effectiveness of this approach, we developed a comprehensive dataset that includes various robotic tasks and failure scenarios. Through extensive experimentation, STAR demonstrated an 86% task planning accuracy and 78% recovery success rate, showing significant improvements over baseline methods. The framework's ability to continuously learn from experience while maintaining structured knowledge representation makes it particularly suitable for long-term deployment in real-world applications.

cs.RO↗

arXiv2025

Inner Thinking Transformer: Leveraging Dynamic Depth Scaling to Foster Adaptive Internal Thinking

Large language models (LLMs) face inherent performance bottlenecks under parameter constraints, particularly in processing critical tokens that demand complex reasoning. Empirical analysis reveals challenging tokens induce abrupt gradient spikes across layers, exposing architectural stress points in standard Transformers. Building on this insight, we propose Inner Thinking Transformer (ITT), which reimagines layer computations as implicit thinking steps. ITT dynamically allocates computation through Adaptive Token Routing, iteratively refines representations via Residual Thinking Connections, and distinguishes reasoning phases using Thinking Step Encoding. ITT enables deeper processing of critical tokens without parameter expansion. Evaluations across 162M-466M parameter models show ITT achieves 96.5\% performance of a 466M Transformer using only 162M parameters, reduces training data by 43.2\%, and outperforms Transformer/Loop variants in 11 benchmarks. By enabling elastic computation allocation during inference, ITT balances performance and efficiency through architecture-aware optimization of implicit thinking pathways.

cs.CL↗

arXiv2025

MA-RLHF: Reinforcement Learning from Human Feedback with Macro Actions

Reinforcement learning from human feedback (RLHF) has demonstrated effectiveness in aligning large language models (LLMs) with human preferences. However, token-level RLHF suffers from the credit assignment problem over long sequences, where delayed rewards make it challenging for the model to discern which actions contributed to preferred outcomes. This hinders learning efficiency and slows convergence.In this paper, we propose MA-RLHF, a simple yet effective RLHF framework that incorporates macro actions -- sequences of tokens or higher-level language constructs -- into the learning process. By operating at higher level of abstraction, our approach reduces the temporal distance between actions and rewards, facilitating faster and more accurate credit assignment. This results in more stable policy gradient estimates and enhances learning efficiency within each episode, all without increasing computational complexity during training or inference. We validate our approach through extensive experiments across various model sizes and tasks, including text summarization, dialogue generation, question answering, and program synthesis. Our method achieves substantial performance improvements over standard RLHF, with performance gains of up to 30% in text summarization and code generation, 18% in dialogue, and 8% in question answering tasks. Notably, our approach reaches parity with vanilla RLHF 1.7 ~ 2 times faster in terms of training time and continues to outperform it with further training. We make our code and data publicly available at https://github.com/ernie-research/MA-RLHF.

cs.CL↗

arXiv2025

Spin frustration and unconventional spin twisting state in van der Waals ferromagnet/antiferromagnet heterostructures

Atomically flat surfaces of van der Waals (vdW) materials pave an avenue for addressing a long-standing fundamental issue of how a perfectly compensated antiferromagnet (AFM) surface frustrates a ferromagnetic (FM) overlayer in FM/AFM heterostructures. By revealing the AFM and FM spin structures separately in vdW Fe5GeTe2/NiPS3 heterostructures, we find that C-type in-plane AFM NiPS3 develops three equivalent AFM domains which are robust against external magnetic field and magnetic coupling with Fe5GeTe2. Consequently, spin frustration at the Fe5GeTe2/NiPS3 interface was shown to develop a perpendicular Fe5GeTe2 magnetization in the interfacial region that switches separately from the bulk of the Fe5GeTe2 magnetizations. In particular, we discover an unconventional spin twisting state that the Fe5GeTe2 spins twist from perpendicular direction near the interface to in-plane direction away from the interface in Fe5GeTe2/NiPS3. Our finding of the twisting spin texture is a unique property of spin frustration in van der Waals magnetic heterostructures.

cond-mat.mtrl-sci↗

arXiv2025

A new perspective on brain stimulation interventions: Optimal stochastic tracking control of brain network dynamics

Network control theory (NCT) has recently been utilized in neuroscience to facilitate our understanding of brain stimulation effects. A particularly useful branch of NCT is optimal control, which focuses on applying theoretical and computational principles of control theory to design optimal strategies to achieve specific goals in neural processes. However, most existing research focuses on optimally controlling brain network dynamics from the original state to a target state at a specific time point. In this paper, we present the first investigation of introducing optimal stochastic tracking control strategy to synchronize the dynamics of the brain network to a target dynamics rather than to a target state at a specific time point. We utilized fMRI data from healthy groups, and cases of stroke and post-stroke aphasia. For all participants, we utilized a gradient descent optimization method to estimate the parameters for the brain network dynamic system. We then utilized optimal stochastic tracking control techniques to drive original unhealthy dynamics by controlling a certain number of nodes to synchronize with target healthy dynamics. Results show that the energy associated with optimal stochastic tracking control is negatively correlated with the intrinsic average controllability of the brain network system, while the energy of the optimal state approaching control is significantly related to the target state value. For a 100-dimensional brain network system, controlling the five nodes with the lowest tracking energy can achieve relatively acceptable dynamics control effects. Our results suggest that stochastic tracking control is more aligned with the objective of brain stimulation interventions, and is closely related to the intrinsic characteristics of the brain network system, potentially representing a new direction for future brain network optimal control research.

q-bio.OT↗

arXiv2025

Graph Structure Learning for Spatial-Temporal Imputation: Adapting to Node and Feature Scales

Spatial-temporal data collected across different geographic locations often suffer from missing values, posing challenges to data analysis. Existing methods primarily leverage fixed spatial graphs to impute missing values, which implicitly assume that the spatial relationship is roughly the same for all features across different locations. However, they may overlook the different spatial relationships of diverse features recorded by sensors in different locations. To address this, we introduce the multi-scale Graph Structure Learning framework for spatial-temporal Imputation (GSLI) that dynamically adapts to the heterogeneous spatial correlations. Our framework encompasses node-scale graph structure learning to cater to the distinct global spatial correlations of different features, and feature-scale graph structure learning to unveil common spatial correlation across features within all stations. Integrated with prominence modeling, our framework emphasizes nodes and features with greater significance in the imputation process. Furthermore, GSLI incorporates cross-feature and cross-temporal representation learning to capture spatial-temporal dependencies. Evaluated on six real incomplete spatial-temporal datasets, GSLI showcases the improvement in data imputation.

cs.LG↗

arXiv2025

Test-Time Training on Video Streams

Prior work has established Test-Time Training (TTT) as a general framework to further improve a trained model at test time. Before making a prediction on each test instance, the model is first trained on the same instance using a self-supervised task such as reconstruction. We extend TTT to the streaming setting, where multiple test instances - video frames in our case - arrive in temporal order. Our extension is online TTT: The current model is initialized from the previous model, then trained on the current frame and a small window of frames immediately before. Online TTT significantly outperforms the fixed-model baseline for four tasks, on three real-world datasets. The improvements are more than 2.2x and 1.5x for instance and panoptic segmentation. Surprisingly, online TTT also outperforms its offline variant that accesses strictly more information, training on all frames from the entire test video regardless of temporal order. This finding challenges those in prior work using synthetic videos. We formalize a notion of locality as the advantage of online over offline TTT, and analyze its role with ablations and a theory based on bias-variance trade-off.

cs.CV↗

arXiv2025

Photoacoustic Iterative Optimization Algorithm with Shape Prior Regularization

Photoacoustic imaging (PAI) suffers from inherent limitations that can degrade the quality of reconstructed results, such as noise, artifacts and incomplete data acquisition caused by sparse sampling or partial array detection. In this study, we proposed a new optimization method for both two-dimensional (2D) and three-dimensional (3D) PAI reconstruction results, called the regularized iteration method with shape prior. The shape prior is a probability matrix derived from the reconstruction results of multiple sets of random partial array signals in a computational imaging system using any reconstruction algorithm, such as Delay-and-Sum (DAS) and Back-Projection (BP). In the probability matrix, high-probability locations indicate high consistency among multiple reconstruction results at those positions, suggesting a high likelihood of representing the true imaging results. In contrast, low-probability locations indicate higher randomness, leaning more towards noise or artifacts. As a shape prior, this probability matrix guides the iteration and regularization of the entire array signal reconstruction results using the original reconstruction algorithm (the same algorithm for processing random partial array signals). The method takes advantage of the property that the similarity of the object to be imitated is higher than that of noise or artifact in the results reconstructed by multiple sets of random partial array signals of the entire imaging system. The probability matrix is taken as a prerequisite for improving the original reconstruction results, and the optimizer is used to further iterate the imaging results to remove noise and artifacts and improve the imaging fidelity. Especially in the case involving sparse view which brings more artifacts, the effect is remarkable. Simulation and real experiments have both demonstrated the superiority of this method.

physics.optics↗

arXiv2024

Mixture of Hidden-Dimensions Transformer

Transformer models encounter challenges in scaling hidden dimensions efficiently, as uniformly increasing them inflates computational and memory costs while failing to emphasize the most relevant features for each token. For further understanding, we study hidden dimension sparsity and observe that trained Transformers utilize only a small fraction of token dimensions, revealing an "activation flow" pattern. Notably, there are shared sub-dimensions with sustained activation across multiple consecutive tokens and specialized sub-dimensions uniquely activated for each token. To better model token-relevant sub-dimensions, we propose MoHD (Mixture of Hidden Dimensions), a sparse conditional activation architecture. Particularly, MoHD employs shared sub-dimensions for common token features and a routing mechanism to dynamically activate specialized sub-dimensions. To mitigate potential information loss from sparsity, we design activation scaling and group fusion mechanisms to preserve activation flow. In this way, MoHD expands hidden dimensions with negligible increases in computation or parameters, efficient training and inference while maintaining performance. Evaluations across 10 NLP tasks show that MoHD surpasses Vanilla Transformers in parameter efficiency and task performance. It achieves 1.7% higher performance with 50% fewer activation parameters and 3.7% higher performance with a 3x parameter expansion at constant activation cost. MOHD offers a new perspective for scaling the model, showcasing the potential of hidden dimension sparsity to boost efficiency

cs.CL↗

arXiv2024

DHA: Learning Decoupled-Head Attention from Transformer Checkpoints via Adaptive Heads Fusion

Large language models (LLMs) with billions of parameters demonstrate impressive performance. However, the widely used Multi-Head Attention (MHA) in LLMs incurs substantial computational and memory costs during inference. While some efforts have optimized attention mechanisms by pruning heads or sharing parameters among heads, these methods often lead to performance degradation or necessitate substantial continued pre-training costs to restore performance. Based on the analysis of attention redundancy, we design a Decoupled-Head Attention (DHA) mechanism. DHA adaptively configures group sharing for key heads and value heads across various layers, achieving a better balance between performance and efficiency. Inspired by the observation of clustering similar heads, we propose to progressively transform the MHA checkpoint into the DHA model through linear fusion of similar head parameters step by step, retaining the parametric knowledge of the MHA checkpoint. We construct DHA models by transforming various scales of MHA checkpoints given target head budgets. Our experiments show that DHA remarkably requires a mere 0.25\% of the original model's pre-training budgets to achieve 97.6\% of performance while saving 75\% of KV cache. Compared to Group-Query Attention (GQA), DHA achieves a 5$\times$ training acceleration, a maximum of 13.93\% performance improvement under 0.01\% pre-training budget, and 4\% relative improvement under 0.05\% pre-training budget.

cs.LG↗