Searcharxiv⌕ Search

arXiv subjects

Haoran Sun

Publications and source records attributed to Haoran Sun.

At least 91 records · Page 5Linked to original sources
arXiv2025

CX-Mind: A Pioneering Multimodal Large Language Model for Interleaved Reasoning in Chest X-ray via Curriculum-Guided Reinforcement Learning

Chest X-ray (CXR) imaging is one of the most widely used diagnostic modalities in clinical practice, encompassing a broad spectrum of diagnostic tasks. Recent advancements have seen the extensive application of reasoning-based multimodal large language models (MLLMs) in medical imaging to enhance diagnostic efficiency and interpretability. However, existing multimodal models predominantly rely on "one-time" diagnostic approaches, lacking verifiable supervision of the reasoning process. This leads to challenges in multi-task CXR diagnosis, including lengthy reasoning, sparse rewards, and frequent hallucinations. To address these issues, we propose CX-Mind, the first generative model to achieve interleaved "think-answer" reasoning for CXR tasks, driven by curriculum-based reinforcement learning and verifiable process rewards (CuRL-VPR). Specifically, we constructed an instruction-tuning dataset, CX-Set, comprising 708,473 images and 2,619,148 samples, and generated 42,828 high-quality interleaved reasoning data points supervised by clinical reports. Optimization was conducted in two stages under the Group Relative Policy Optimization framework: initially stabilizing basic reasoning with closed-domain tasks, followed by transfer to open-domain diagnostics, incorporating rule-based conditional process rewards to bypass the need for pretrained reward models. Extensive experimental results demonstrate that CX-Mind significantly outperforms existing medical and general-domain MLLMs in visual understanding, text generation, and spatiotemporal alignment, achieving an average performance improvement of 25.1% over comparable CXR-specific models. On real-world clinical dataset (Rui-CXR), CX-Mind achieves a mean recall@1 across 14 diseases that substantially surpasses the second-best results, with multi-center expert evaluations further confirming its clinical utility across multiple dimensions.

cs.LG↗
arXiv2025

EnTao-GPM: DNA Foundation Model for Predicting the Germline Pathogenic Mutations

Distinguishing pathogenic mutations from benign polymorphisms remains a critical challenge in precision medicine. EnTao-GPM, developed by Fudan University and BioMap, addresses this through three innovations: (1) Cross-species targeted pre-training on disease-relevant mammalian genomes (human, pig, mouse), leveraging evolutionary conservation to enhance interpretation of pathogenic motifs, particularly in non-coding regions; (2) Germline mutation specialization via fine-tuning on ClinVar and HGMD, improving accuracy for both SNVs and non-SNVs; (3) Interpretable clinical framework integrating DNA sequence embeddings with LLM-based statistical explanations to provide actionable insights. Validated against ClinVar, EnTao-GPM demonstrates superior accuracy in mutation classification. It revolutionizes genetic testing by enabling faster, more accurate, and accessible interpretation for clinical diagnostics (e.g., variant assessment, risk identification, personalized treatment) and research, advancing personalized medicine.

q-bio.GN↗
arXiv2025

An Uncertainty-Driven Adaptive Self-Alignment Framework for Large Language Models

Large Language Models (LLMs) have demonstrated remarkable progress in instruction following and general-purpose reasoning. However, achieving high-quality alignment with human intent and safety norms without human annotations remains a fundamental challenge. In this work, we propose an Uncertainty-Driven Adaptive Self-Alignment (UDASA) framework designed to improve LLM alignment in a fully automated manner. UDASA first generates multiple responses for each input and quantifies output uncertainty across three dimensions: semantics, factuality, and value alignment. Based on these uncertainty scores, the framework constructs preference pairs and categorizes training samples into three stages, conservative, moderate, and exploratory, according to their uncertainty difference. The model is then optimized progressively across these stages. In addition, we conduct a series of preliminary studies to validate the core design assumptions and provide strong empirical motivation for the proposed framework. Experimental results show that UDASA outperforms existing alignment methods across multiple tasks, including harmlessness, helpfulness, truthfulness, and controlled sentiment generation, significantly improving model performance.

cs.AI↗
arXiv2025

Hierarchical Memory for High-Efficiency Long-Term Reasoning in LLM Agents

Long-term memory is one of the key factors influencing the reasoning capabilities of Large Language Model Agents (LLM Agents). Incorporating a memory mechanism that effectively integrates past interactions can significantly enhance decision-making and contextual coherence of LLM Agents. While recent works have made progress in memory storage and retrieval, such as encoding memory into dense vectors for similarity-based search or organizing knowledge in the form of graph, these approaches often fall short in structured memory organization and efficient retrieval. To address these limitations, we propose a Hierarchical Memory (H-MEM) architecture for LLM Agents that organizes and updates memory in a multi-level fashion based on the degree of semantic abstraction. Each memory vector is embedded with a positional index encoding pointing to its semantically related sub-memories in the next layer. During the reasoning phase, an index-based routing mechanism enables efficient, layer-by-layer retrieval without performing exhaustive similarity computations. We evaluate our method on five task settings from the LoCoMo dataset. Experimental results show that our approach consistently outperforms five baseline methods, demonstrating its effectiveness in long-term dialogue scenarios.

cs.CL↗
arXiv2025

A Novel Self-Evolution Framework for Large Language Models

The capabilities of Large Language Models (LLMs) are limited to some extent by pre-training, so some researchers optimize LLMs through post-training. Existing post-training strategies, such as memory-based retrieval or preference optimization, improve user alignment yet fail to enhance the model's domain cognition. To bridge this gap, we propose a novel Dual-Phase Self-Evolution (DPSE) framework that jointly optimizes user preference adaptation and domain-specific competence. DPSE introduces a Censor module to extract multi-dimensional interaction signals and estimate satisfaction scores, which guide structured data expansion via topic-aware and preference-driven strategies. These expanded datasets support a two-stage fine-tuning pipeline: supervised domain grounding followed by frequency-aware preference optimization. Experiments across general NLP benchmarks and long-term dialogue tasks demonstrate that DPSE consistently outperforms Supervised Fine-Tuning, Preference Optimization, and Memory-Augmented baselines. Ablation studies validate the contribution of each module. In this way, our framework provides an autonomous path toward continual self-evolution of LLMs.

cs.CL↗
arXiv2025

DatasetAgent: A Novel Multi-Agent System for Auto-Constructing Datasets from Real-World Images

Common knowledge indicates that the process of constructing image datasets usually depends on the time-intensive and inefficient method of manual collection and annotation. Large models offer a solution via data generation. Nonetheless, real-world data are obviously more valuable comparing to artificially intelligence generated data, particularly in constructing image datasets. For this reason, we propose a novel method for auto-constructing datasets from real-world images by a multiagent collaborative system, named as DatasetAgent. By coordinating four different agents equipped with Multi-modal Large Language Models (MLLMs), as well as a tool package for image optimization, DatasetAgent is able to construct high-quality image datasets according to user-specified requirements. In particular, two types of experiments are conducted, including expanding existing datasets and creating new ones from scratch, on a variety of open-source datasets. In both cases, multiple image datasets constructed by DatasetAgent are used to train various vision models for image classification, object detection, and image segmentation.

cs.CV↗
arXiv2025

Introspection of Thought Helps AI Agents

AI Agents rely on Large Language Models (LLMs) and Multimodal-LLMs (MLLMs) to perform interpretation and inference in text and image tasks without post-training, where LLMs and MLLMs play the most critical role and determine the initial ability and limitations of AI Agents. Usually, AI Agents utilize sophisticated prompt engineering and external reasoning framework to obtain a promising interaction with LLMs, e.g., Chain-of-Thought, Iteration of Thought and Image-of-Thought. However, they are still constrained by the inherent limitations of LLM in understanding natural language, and the iterative reasoning process will generate a large amount of inference cost. To this end, we propose a novel AI Agent Reasoning Framework with Introspection of Thought (INoT) by designing a new LLM-Read code in prompt. It enables LLM to execute programmatic dialogue reasoning processes following the code in prompt. Therefore, self-denial and reflection occur within LLM instead of outside LLM, which can reduce token cost effectively. Through our experiments on six benchmarks for three different tasks, the effectiveness of INoT is verified, with an average improvement of 7.95\% in performance, exceeding the baselines. Furthermore, the token cost of INoT is lower on average than the best performing method at baseline by 58.3\%. In addition, we demonstrate the versatility of INoT in image interpretation and inference through verification experiments.

cs.AI↗
arXiv2025

Strong Temperature Dependence of Thermal Conductivity in High-Purity Cubic Boron Arsenide

Materials with high thermal conductivity are needed to conduct heat away from hot spots in high power electronics and optoelectronic devices. Cubic boron arsenide (c-BAs) has a high thermal conductivity due to its special phonon dispersion relation. Previous experimental studies of c-BAs report a room-temperature thermal conductivity between 1000 and 1300 W m-1 K-1. We synthesized high purity isotopically enriched c-BAs single crystals with room-temperature thermal conductivity of around 1500 W m-1 K-1. Using time-domain thermoreflectance (TDTR), we measured thermal conductivity and found a 1/T2 temperature dependence between 300 K and 600 K - slightly stronger than predictions from state-of-the-art theoretical models. Brillouin and Raman scattering revealed minimal changes in phonon frequencies over the same temperature range, suggesting that the observed 1/T2 dependence is not caused by temperature dependent changes in phonon dispersion. To probe defect densities in the BAs crystals we studied, we conducted transient reflectivity microscopy (TRM) measurements of absorption at sub-bandgap photon energies. We observe a correlation between TRM signal intensity and thermal conductivity. Notably, samples with thermal conductivity near 1500 W m-1 K-1 still exhibited nonzero TRM signals, suggesting the presence of defects despite the high thermal conductivity.

cond-mat.mtrl-sci↗
arXiv2025

Core Knowledge Deficits in Multi-Modal Language Models

While Multi-modal Large Language Models (MLLMs) demonstrate impressive abilities over high-level perception and reasoning, their robustness in the wild remains limited, often falling short on tasks that are intuitive and effortless for humans. We examine the hypothesis that these deficiencies stem from the absence of core knowledge--rudimentary cognitive abilities innate to humans from early childhood. To explore the core knowledge representation in MLLMs, we introduce CoreCognition, a large-scale benchmark encompassing 12 core knowledge concepts grounded in developmental cognitive science. We evaluate 230 models with 11 different prompts, leading to a total of 2,530 data points for analysis. Our experiments uncover four key findings, collectively demonstrating core knowledge deficits in MLLMs: they consistently underperform and show reduced, or even absent, scalability on low-level abilities relative to high-level ones. Finally, we propose Concept Hacking, a novel controlled evaluation method that reveals MLLMs fail to progress toward genuine core knowledge understanding, but instead rely on shortcut learning as they scale.

cs.CL↗
arXiv2025

Amulet: ReAlignment During Test Time for Personalized Preference Adaptation of LLMs

How to align large language models (LLMs) with user preferences from a static general dataset has been frequently studied. However, user preferences are usually personalized, changing, and diverse regarding culture, values, or time. This leads to the problem that the actual user preferences often do not coincide with those trained by the model developers in the practical use of LLMs. Since we cannot collect enough data and retrain for every demand, researching efficient real-time preference adaptation methods based on the backbone LLMs during test time is important. To this end, we introduce Amulet, a novel, training-free framework that formulates the decoding process of every token as a separate online learning problem with the guidance of simple user-provided prompts, thus enabling real-time optimization to satisfy users' personalized preferences. To reduce the computational cost brought by this optimization process for each token, we additionally provide a closed-form solution for each iteration step of the optimization process, thereby reducing the computational time cost to a negligible level. The detailed experimental results demonstrate that Amulet can achieve significant performance improvements in rich settings with combinations of different LLMs, datasets, and user preferences, while maintaining acceptable computational efficiency.

cs.CL↗
arXiv2025

Curiosity-Driven Reinforcement Learning from Human Feedback

Reinforcement learning from human feedback (RLHF) has proven effective in aligning large language models (LLMs) with human preferences, but often at the cost of reduced output diversity. This trade-off between diversity and alignment quality remains a significant challenge. Drawing inspiration from curiosity-driven exploration in reinforcement learning, we introduce curiosity-driven RLHF (CD-RLHF), a framework that incorporates intrinsic rewards for novel states, alongside traditional sparse extrinsic rewards, to optimize both output diversity and alignment quality. We demonstrate the effectiveness of CD-RLHF through extensive experiments on a range of tasks, including text summarization and instruction following. Our approach achieves significant gains in diversity on multiple diversity-oriented metrics while maintaining alignment with human preferences comparable to standard RLHF. We make our code publicly available at https://github.com/ernie-research/CD-RLHF.

cs.CL↗
arXiv2025

NTIRE 2025 Challenge on Day and Night Raindrop Removal for Dual-Focused Images: Methods and Results

This paper reviews the NTIRE 2025 Challenge on Day and Night Raindrop Removal for Dual-Focused Images. This challenge received a wide range of impressive solutions, which are developed and evaluated using our collected real-world Raindrop Clarity dataset. Unlike existing deraining datasets, our Raindrop Clarity dataset is more diverse and challenging in degradation types and contents, which includes day raindrop-focused, day background-focused, night raindrop-focused, and night background-focused degradations. This dataset is divided into three subsets for competition: 14,139 images for training, 240 images for validation, and 731 images for testing. The primary objective of this challenge is to establish a new and powerful benchmark for the task of removing raindrops under varying lighting and focus conditions. There are a total of 361 participants in the competition, and 32 teams submitting valid solutions and fact sheets for the final testing phase. These submissions achieved state-of-the-art (SOTA) performance on the Raindrop Clarity dataset. The project can be found at https://lixinustc.github.io/CVPR-NTIRE2025-RainDrop-Competition.github.io/.

cs.CV↗
arXiv2025

NTIRE 2025 Challenge on Cross-Domain Few-Shot Object Detection: Methods and Results

Cross-Domain Few-Shot Object Detection (CD-FSOD) poses significant challenges to existing object detection and few-shot detection models when applied across domains. In conjunction with NTIRE 2025, we organized the 1st CD-FSOD Challenge, aiming to advance the performance of current object detectors on entirely novel target domains with only limited labeled data. The challenge attracted 152 registered participants, received submissions from 42 teams, and concluded with 13 teams making valid final submissions. Participants approached the task from diverse perspectives, proposing novel models that achieved new state-of-the-art (SOTA) results under both open-source and closed-source settings. In this report, we present an overview of the 1st NTIRE 2025 CD-FSOD Challenge, highlighting the proposed solutions and summarizing the results submitted by the participants.

cs.CV↗
arXiv2025

LCV2I: Communication-Efficient and High-Performance Collaborative Perception Framework with Low-Resolution LiDAR

Vehicle-to-Infrastructure (V2I) collaborative perception leverages data collected by infrastructure's sensors to enhance vehicle perceptual capabilities. LiDAR, as a commonly used sensor in cooperative perception, is widely equipped in intelligent vehicles and infrastructure. However, its superior performance comes with a correspondingly high cost. To achieve low-cost V2I, reducing the cost of LiDAR is crucial. Therefore, we study adopting low-resolution LiDAR on the vehicle to minimize cost as much as possible. However, simply reducing the resolution of vehicle's LiDAR results in sparse point clouds, making distant small objects even more blurred. Additionally, traditional communication methods have relatively low bandwidth utilization efficiency. These factors pose challenges for us. To balance cost and perceptual accuracy, we propose a new collaborative perception framework, namely LCV2I. LCV2I uses data collected from cameras and low-resolution LiDAR as input. It also employs feature offset correction modules and regional feature enhancement algorithms to improve feature representation. Finally, we use regional difference map and regional score map to assess the value of collaboration content, thereby improving communication bandwidth efficiency. In summary, our approach achieves high perceptual performance while substantially reducing the demand for high-resolution sensors on the vehicle. To evaluate this algorithm, we conduct 3D object detection in the real-world scenario of DAIR-V2X, demonstrating that the performance of LCV2I consistently surpasses currently existing algorithms.

cs.CV↗
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↗
arXiv2024

Language-based Audio Retrieval with Co-Attention Networks

In recent years, user-generated audio content has proliferated across various media platforms, creating a growing need for efficient retrieval methods that allow users to search for audio clips using natural language queries. This task, known as language-based audio retrieval, presents significant challenges due to the complexity of learning semantic representations from heterogeneous data across both text and audio modalities. In this work, we introduce a novel framework for the language-based audio retrieval task that leverages co-attention mechanismto jointly learn meaningful representations from both modalities. To enhance the model's ability to capture fine-grained cross-modal interactions, we propose a cascaded co-attention architecture, where co-attention modules are stacked or iterated to progressively refine the semantic alignment between text and audio. Experiments conducted on two public datasets show that the proposed method can achieve better performance than the state-of-the-art method. Specifically, our best performed co-attention model achieves a 16.6% improvement in mean Average Precision on Clotho dataset, and a 15.1% improvement on AudioCaps.

cs.SD↗
arXiv2024

Control of open quantum systems: Manipulation of a qubit coupled to a thermal bath by an external driving field

Fast and reliable manipulation with qubits is fundamental for any quantum technology. The implementation of these manipulations in physical systems is the focus of studies involving optimal control theory. Realistic physical devices are open quantum systems. So far, studies in optimal control theory have primarily utilized the Redfield/Lindblad quantum master equation to simulate the dynamics of such systems. However, this Markov description is not always sufficient. Here, we present a study of qubit control utilizing the nonequilibrium Green's function method. We compare the traditional master equation with more general Green's function results and demonstrate that even in the parameter regime suitable for the application of the Redfield/Lindblad approach, the two methods yield drastically different results when addressing evolution involving mixed states. In particular, we find that, in addition to predicting different optimal driving profiles, a more accurate description of system evolution enables the system to reach the desired final state much more quickly. We argue that the primary reason for this is the significance of the non-Markov description of driven system dynamics due to the effect of time-dependent driving on dissipation.

quant-ph↗