arXiv subjects

Furong Li

Publications and source records attributed to Furong Li.

6 recordsLinked to original sources

arXiv2026

Two kinds of Galactic source populations could explain the cosmic-ray observation up to the "knee" region

Observations of diffuse gamma rays above hundreds of TeV from the Galactic disk provide strong evidence for the existence of PeV cosmic-ray accelerators--so-called PeVatrons--in the Galaxy. However, mounting observations have ruled out most supernova remnants as likely PeVatron candidates, suggesting instead that multiple populations of cosmic-ray sources exist in the Galaxy. Recently, the LHAASO collaboration reported the detection of ultra-high-energy gamma rays from microquasars, establishing that the black holes in these systems, which accrete matter from companion stars, are powerful PeV particle accelerators. In this work, we propose a two-component source model to explain the observed cosmic-ray spectra and composition up to the PeV range. Below approximately 100 TeV, supernova remnants serve as the dominant sources; above this energy, microquasars are considered the primary candidate population. Within this scenario, the assumption of a charge-dependent cutoff well accounts for the latest measurements, including the proton and helium spectra up to the PeV range, the energy-dependent composition, and the all-particle spectrum. In contrast, the nuclei-dependent cutoff hypothesis is ruled out by the data.

astro-ph.HE↗

arXiv2025

Medusa: Cross-Modal Transferable Adversarial Attacks on Multimodal Medical Retrieval-Augmented Generation

With the rapid advancement of retrieval-augmented vision-language models, multimodal medical retrieval-augmented generation (MMed-RAG) systems are increasingly adopted in clinical decision support. These systems enhance medical applications by performing cross-modal retrieval to integrate relevant visual and textual evidence for tasks, e.g., report generation and disease diagnosis. However, their complex architecture also introduces underexplored adversarial vulnerabilities, particularly via visual input perturbations. In this paper, we propose Medusa, a novel framework for crafting cross-modal transferable adversarial attacks on MMed-RAG systems under a black-box setting. Specifically, Medusa formulates the attack as a perturbation optimization problem, leveraging a multi-positive InfoNCE loss (MPIL) to align adversarial visual embeddings with medically plausible but malicious textual targets, thereby hijacking the retrieval process. To enhance transferability, we adopt a surrogate model ensemble and design a dual-loop optimization strategy augmented with invariant risk minimization (IRM). Extensive experiments on two real-world medical tasks, including medical report generation and disease diagnosis, demonstrate that Medusa achieves over 90% average attack success rate across various generation models and retrievers under appropriate parameter configuration, while remaining robust against four mainstream defenses, outperforming state-of-the-art baselines. Our results reveal critical vulnerabilities in the MMed-RAG systems and highlight the necessity of robustness benchmarking in safety-critical medical applications. The code and data are available at https://anonymous.4open.science/r/MMed-RAG-Attack-F05A.

cs.CR↗

arXiv2025

Structured Semantics from Unstructured Notes: Language Model Approaches to EHR-Based Decision Support

The advent of large language models (LLMs) has opened new avenues for analyzing complex, unstructured data, particularly within the medical domain. Electronic Health Records (EHRs) contain a wealth of information in various formats, including free text clinical notes, structured lab results, and diagnostic codes. This paper explores the application of advanced language models to leverage these diverse data sources for improved clinical decision support. We will discuss how text-based features, often overlooked in traditional high dimensional EHR analysis, can provide semantically rich representations and aid in harmonizing data across different institutions. Furthermore, we delve into the challenges and opportunities of incorporating medical codes and ensuring the generalizability and fairness of AI models in healthcare.

cs.IR↗

arXiv2025

MFM-DA: Instance-Aware Adaptor and Hierarchical Alignment for Efficient Domain Adaptation in Medical Foundation Models

Medical Foundation Models (MFMs), trained on large-scale datasets, have demonstrated superior performance across various tasks. However, these models still struggle with domain gaps in practical applications. Specifically, even after fine-tuning on source-domain data, task-adapted foundation models often perform poorly in the target domain. To address this challenge, we propose a few-shot unsupervised domain adaptation (UDA) framework for MFMs, named MFM-DA, which only leverages a limited number of unlabeled target-domain images. Our approach begins by training a Denoising Diffusion Probabilistic Model (DDPM), which is then adapted to the target domain using a proposed dynamic instance-aware adaptor and a distribution direction loss, enabling the DDPM to translate source-domain images into the target domain style. The adapted images are subsequently processed through the MFM, where we introduce a designed channel-spatial alignment Low-Rank Adaptation (LoRA) to ensure effective feature alignment. Extensive experiments on optic cup and disc segmentation tasks demonstrate that MFM-DA outperforms state-of-the-art methods. Our work provides a practical solution to the domain gap issue in real-world MFM deployment. Code will be available at here.

cs.CV↗

arXiv2019

Ex-ante dynamic network tariffs for transmission cost recovery

This paper proposes a novel tariff scheme and a new optimization framework in order to address the recovery of fixed investment costs in transmission network planning, particularly against rising demand elasticity. At the moment, ex-post network tariffs are utilized in addition to congestion revenues to fully recover network costs, which often leads to over/under fixed cost recovery, thus increasing the investment risk. Furthermore, in the case of agents with elastic market curves, ex-post tariffs can cause several inefficiencies, such as mistrustful bidding to exploit ex-post schemes, imperfect information in applied costs and cleared quantities, and negative surplus for marginal generators and consumers. These problems are exacerbated by the increasing price-elasticity of demand, caused for example by the diffusion of demand response technologies. To address these issues, we design a dynamic ex-ante tariff scheme that explicitly accounts for the effect of tariffs in the longterm network planning problem and in the underlying market clearing process. Using linearization techniques and a novel reformulation of the congestion rent, the long-term network planning problem is reformulated as a single mixed-integer linear problem which returns the combined optimal values of network expansion and associated tariffs, while accounting for price-elastic agents and lumpy investments. The advantages of the proposed approach in terms of cost recovery, market equilibrium and increased social welfare are discussed qualitatively and are validated in numerical case studies.

eess.SY↗

arXiv2017

Discovering Multiple Truths with a Hybrid Model

Many data management applications require integrating information from multiple sources. The sources may not be accurate and provide erroneous values. We thus have to identify the true values from conflicting observations made by the sources. The problem is further complicated when there may exist multiple truths (e.g., a book written by several authors). In this paper we propose a model called Hybrid that jointly makes two decisions: how many truths there are, and what they are. It considers the conflicts between values as important evidence for ruling out wrong values, while keeps the flexibility of allowing multiple truths. In this way, Hybrid is able to achieve both high precision and high recall.

cs.DB↗