arXiv subjects

Frank Li

Publications and source records attributed to Frank Li.

At least 19 recordsLinked to original sources

arXiv2026

Grid Trouble in Paradise: Uncovering Vulnerable Distributed Energy Resources and Their Grid-Level Risks

Grid-connected solar distributed energy resources (DERs), such as solar inverters and monitoring platforms, have been deployed at unprecedented scale over the past few years, with global solar capacity more than doubling since 2022. To support monitoring and control, many of these systems are Internet-connected and configured by installers or end users, yet the real-world scale of their Internet exposure and the implications for power grid operation remain poorly understood. In this paper, we present an Internet-scale evaluation of exposed and vulnerable solar DER infrastructure, and assess the risk that compromised DERs can pose to energy grids. We develop a method for accurately identifying solar DERs from Internet scanning data, and discover a diverse population of over 66,000 Internet-exposed solar DERs. We detect that at least 10,000 of these DERs may have known CVEs, such as unauthenticated monitoring and control endpoints. To assess the risk that these vulnerable DERs pose to a power grid, we use an electric grid network for Oahu, Hawaii, established and used by the power system research community, and conduct a power system analysis. Our evaluation shows that by compromising exposed DERs, attackers can cause voltage and line flow violations across multiple locations in the Oahu network, resulting in a range of consequences from degraded power quality to damaged power system components to power outages. Ultimately, our work brings to light the emerging threat of grid-connected DERs, and provides directions for improving energy security.

cs.CR↗

arXiv2026

Detecting and Characterizing Massively Shared IP Addresses

IP addresses are commonly shared across devices and users for a variety of reasons, including NAT and proxies. These technologies operate at different scales, from residential NATs that share an IP address across devices in a home to large-scale Carrier Grade NATs that share hundreds or thousands of users on a single IP. Cases of large-scale IP sharing are distinct as they have significant implications for IP-based mechanisms such as attribution, blocklisting, and rate-limiting, where the consequences of mishandling affect a large quantity of end-users and organizations. In this work, we detect and characterize IP addresses shared at large scales, which we coin massively shared. Leveraging diurnal patterns in traffic shape, we use data from a large CDN to characterize these IPs globally. We broadly find that massive IP sharing is responsible for a large fraction of IPv4 traffic, concentrated in a small fraction of address space, with over 40% of total traffic coming from less than 2% of active IP addresses. We observe distinct patterns in deployment geographically, with particularly high rates of massively shared traffic from some smaller countries. Comparatively, in IPv6, we find far fewer massively shared addresses with some surprising exceptions among mobile providers. We additionally contextualize these addresses by other network characteristics, including identifying cellular connectivity and dual-stack capabilities, and identifying several instances of massively shared IPs in proxy services hosted on cloud networks. Finally, we find that rates of massively shared traffic are increasing over time, predicting future reliance on these technologies. Our work contextualizes the state of IP sharing, providing a uniquely broad perspective globally.

cs.NI↗

arXiv2026

Mind the Gap: Policy vs Reality in Post-Quantum TLS Deployment

Post-quantum cryptography (PQC) has evolved from a long-term planning concern into an operational priority. Following NIST's standardization of PQC, governments and standard bodies published transition roadmaps outlining migration timelines, priority sectors, and deployment strategies. However, our survey of these policies reveals substantial divergence in technical prescriptions and urgency. It remains unclear how widely PQC has been adopted in practice and how policy differences translate into observable deployment outcomes. To address this gap, we present the first longitudinal measurement study of post-quantum TLS (PQ-TLS) adoption. By establishing more than 2 billion TLS handshakes, we analyze cryptographic negotiation behavior across 1 million domains from 11 globally distributed vantage points. Despite varied policy guidance, we observe configuration convergence: PQ-TLS deployment overwhelmingly centers on a single hybrid construction, and much of the apparent progress is driven by managed infrastructure providers. National timelines and sectoral priorities show limited correspondence with observed deployment patterns. Contrary to early experimental studies suggesting measurable overhead, we find that PQ-TLS introduces no meaningful latency increase in Internet settings, although it is frequently deployed alongside legacy TLS configurations. Together, these findings highlight a gap between policy expectations and early deployment reality, and provide empirical insight to inform more grounded PQ-TLS transition.

cs.NI↗

arXiv2026

RadHarmony: Radiological Data Handling in the Era of Agentic AI

Training deep learning models on radiological images requires integrating heterogeneous datasets across different sources, file formats, directory layouts, label schemas, and annotation types. We present RadHarmony, an open-source Python library that provides a unified API for loading, harmonizing, and augmenting radiological datasets, with a primary focus on chest radiographs and early support for computed tomography (CT) and magnetic resonance imaging (MRI). RadHarmony standardizes metadata from 24 public datasets into a single tabular format, wraps MONAI's map-style datasets for deep-learning-ready sample delivery with optional on-disk caching, and supports classification labels, segmentation masks, bounding boxes, and radiology report text through a single interface, with an interactive visualization tool for dataset exploration and verification. To lower the barrier for integrating new datasets, RadHarmony introduces an AI-agent skill that guides the full integration workflow from raw data inspection through code generation and testing. We demonstrate the library's utility by pretraining RadHarmony-ViT, a reference vision transformer baseline that combines three heterogeneous chest radiograph datasets with no dataset-specific code. The code and pretrained model weights are available at https://github.com/f10409/RadHarmony.

cs.AI↗

arXiv2026

Frozen Foundation-Model Embeddings Discard Small-Lesion Signal in Chest Radiography: Implications for Pre-Deployment Evaluation

Frozen vision-transformer (ViT) foundation-model embeddings increasingly serve as the substrate for downstream chest-radiography (CXR) pipelines, yet where small-scale, low-contrast signal is retained or lost in the frozen forward pass has not been systematically quantified across architectures, pretraining domains, and objectives. We probed five frozen ViTs (RAD-DINO, DINOv2-B/14, DINOv3 ViT-7B, BiomedCLIP, MedSigLIP) and a frozen DINO-pretrained ResNet-50 architectural control across three large CXR cohorts (NIH-CXR14, MIMIC-CXR, Emory-CXR; aggregate pool n=492,724) and ChestX-Det10 (n=3,543; 1,462 small-lesion bounding boxes across Calcification, Nodule, Mass). Each model was evaluated with a small-scale-perturbation panel and a region-aware bounding-box-stratified probe on real lesions, comparing three pooling modes from the same forward pass: classification token (CLS), patch-mean (mean over all final-layer patch tokens), and bounding-box-restricted patch-local. On the perturbation panel, CLS embeddings sat at the chance floor (area under the ROC curve [AUC] 0.500-0.524); patch-mean was indistinguishable from CLS on iso-blur and reticular-fine cells but rose with CLS on larger directional-blur footprints, while disease AUC on globally decided tasks ranged 0.642-0.913. Patch-local probes recovered AUC ~1.0 from the same forward pass (per-model mean improvement +0.412 to +0.488); the ResNet-50 control reproduced the chance floor. On ChestX-Det10, image-level CLS classification showed within-class small-versus-large stratum gaps up to +0.243 AUC; bounding-box-level patch-local pooling on the same forward pass recovered AUC >= 0.899 on every (model x class) cell. Frozen ViT embeddings silently suppress small-scale signal at the global-aggregation step; the signal is recoverable from patch tokens conditional on a region of interest.

cs.CV↗

arXiv2026

MultiMedVision: Multi-Modal Medical Vision Framework

Multi-modal medical imaging enables comprehensive diagnostics, yet current foundation models process 2D (e.g. X-ray) and 3D (e.g. CT) data with separate, dimensionality-specific architectures. We present MultiMedVision, a unified framework for joint 2D/3D representation learning built on a Sparse Vision Transformer. Our model uses 3D Rotary Positional Embeddings and variable-length sequence packing to process mixed-modality batches natively within a shared latent space, without modality-specific adapters or treating 3D volumes as 2D slice sequences. Trained with a self-supervised objective on chest X-rays (MIMIC-CXR) and CT scans (CT-RATE), and using a single shared encoder with 5x less data, MultiMedVision achieves competitive performance on both 2D benchmarks (Macro AUROC 0.82 on MIMIC, 0.84 on CheXpert) and 3D tasks (0.85 on CT-RATE). Analysis of the learned representations reveals coexisting modality-specific and shared feature subspaces, demonstrating that unified cross-dimensional representation learning is feasible without sacrificing modality-specific performance.

cs.CV↗

arXiv2026

OpenClaw PRISM: A Zero-Fork, Defense-in-Depth Runtime Security Layer for Tool-Augmented LLM Agents

Tool-augmented LLM agents introduce security risks that extend beyond user-input filtering, including indirect prompt injection through fetched content, unsafe tool execution, credential leakage, and tampering with local control files. We present OpenClaw PRISM, a zero-fork runtime security layer for OpenClaw-based agent gateways. PRISM combines an in-process plugin with optional sidecar services and distributes enforcement across ten lifecycle hooks spanning message ingress, prompt construction, tool execution, tool-result persistence, outbound messaging, sub-agent spawning, and gateway startup. Rather than introducing a novel detection model, PRISM integrates a hybrid heuristic-plus-LLM scanning pipeline, conversation- and session-scoped risk accumulation with TTL-based decay, policy-enforced controls over tools, paths, private networks, domain tiers, and outbound secret patterns, and a tamper-evident audit and operations plane with integrity verification and hot-reloadable policy management. We outline an evaluation methodology and benchmark pipeline for measuring security effectiveness, false positives, layer contribution, runtime overhead, and operational recoverability in an agent-runtime setting, and we report current preliminary benchmark results on curated same-slice experiments and operational microbenchmarks. The system targets deployable runtime defense for real agent gateways rather than benchmark-only detection.

cs.CR↗

arXiv2026

DMind-3: A Sovereign Edge--Local--Cloud AI System with Controlled Deliberation and Correction-Based Tuning for Safe, Low-Latency Transaction Execution

This paper introduces DMind-3, a sovereign Edge-Local-Cloud intelligence stack designed to secure irreversible financial execution in Web3 environments against adversarial risks and strict latency constraints. While existing cloud-centric assistants compromise privacy and fail under network congestion, and purely local solutions lack global ecosystem context, DMind-3 resolves these tensions by decomposing capability into three cooperating layers: a deterministic signing-time intent firewall at the edge, a private high-fidelity reasoning engine on user hardware, and a policy-governed global context synthesizer in the cloud. We propose policy-driven selective offloading to route computation based on privacy sensitivity and uncertainty, supported by two novel training objectives: Hierarchical Predictive Synthesis (HPS) for fusing time-varying macro signals, and Contrastive Chain-of-Correction Supervised Fine-Tuning (C$^3$-SFT) to enhance local verification reliability. Extensive evaluations demonstrate that DMind-3 achieves a 93.7% multi-turn success rate in protocol-constrained tasks and superior domain reasoning compared to general-purpose baselines, providing a scalable framework where safety is bound to the edge execution primitive while maintaining sovereignty over sensitive user intent.

cs.CR↗

arXiv2026

Improving Database Performance by Application-side Transaction Merging

This paper explores a new opportunity to improve the performance of transaction processing at the application side by merging structurely similar statements or transactions. Concretely, we re-write transactions to 1) merge similar statements using specific SQL semantics; 2) eliminate redundant reads; and 3) merge contending statements across transactions by pre-computing their aggregated effect. Following this idea, we present the design of TransactionMerger, a middleware to collect and merge transactions across different clients. We further present a static analysis tool to identify the merging opportunity without violating isolation as well as our experience of re-writing transactions in TPC-C and Spree, a popular real-world application. Our evaluation shows that such transaction merging can improve TPC-C throughput by up to 2.65X and Spree throughput by 3.52X.

cs.DB↗

arXiv2025

Feature Quality and Adaptability of Medical Foundation Models: A Comparative Evaluation for Radiographic Classification and Segmentation

Foundation models (FMs) promise to generalize medical imaging, but their effectiveness varies. It remains unclear how pre-training domain (medical vs. general), paradigm (e.g., text-guided), and architecture influence embedding quality, hindering the selection of optimal encoders for specific radiology tasks. To address this, we evaluate vision encoders from eight medical and general-domain FMs for chest X-ray analysis. We benchmark classification (pneumothorax, cardiomegaly) and segmentation (pneumothorax, cardiac boundary) using linear probing and fine-tuning. Our results show that domain-specific pre-training provides a significant advantage; medical FMs consistently outperformed general-domain models in linear probing, establishing superior initial feature quality. However, feature utility is highly task-dependent. Pre-trained embeddings were strong for global classification and segmenting salient anatomy (e.g., heart). In contrast, for segmenting complex, subtle pathologies (e.g., pneumothorax), all FMs performed poorly without significant fine-tuning, revealing a critical gap in localizing subtle disease. Subgroup analysis showed FMs use confounding shortcuts (e.g., chest tubes for pneumothorax) for classification, a strategy that fails for precise segmentation. We also found that expensive text-image alignment is not a prerequisite; image-only (RAD-DINO) and label-supervised (Ark+) FMs were among top performers. Notably, a supervised, end-to-end baseline remained highly competitive, matching or exceeding the best FMs on segmentation tasks. These findings show that while medical pre-training is beneficial, architectural choices (e.g., multi-scale) are critical, and pre-trained features are not universally effective, especially for complex localization tasks where supervised models remain a strong alternative.

cs.CV↗

arXiv2025

A Sea of Cyber Threats: Maritime Cybersecurity from the Perspective of Mariners

Maritime systems, including ships and ports, are critical components of global infrastructure, essential for transporting over 80% of the world's goods and supporting internet connectivity. However, these systems face growing cybersecurity threats, as shown by recent attacks disrupting Maersk, one of the world's largest shipping companies, causing widespread impacts on international trade. The unique challenges of the maritime environment--such as diverse operational conditions, extensive physical access points, fragmented regulatory frameworks, and its deeply interconnected structure--require maritime-specific cybersecurity research. Despite the sector's importance, maritime cybersecurity remains underexplored, leaving significant gaps in understanding its challenges and risks. To address these gaps, we investigate how maritime system operators perceive and navigate cybersecurity challenges within this complex landscape. We conducted a user study comprising surveys and semi-structured interviews with 21 officer-level mariners. Participants reported direct experiences with shipboard cyber-attacks, including GPS spoofing and logistics-disrupting ransomware, demonstrating the real-world impact of these threats. Our findings reveal systemic and human-centric issues, such as training poorly aligned with maritime needs, insufficient detection and response tools, and serious gaps in mariners' cybersecurity understanding. Our contributions include a categorization of threats identified by mariners and recommendations for improving maritime security, including better training, response protocols, and regulation. These insights aim to guide future research and policy to strengthen the resilience of maritime systems.

cs.CR↗

arXiv2025

Impact of Label Noise from Large Language Models Generated Annotations on Evaluation of Diagnostic Model Performance

Large language models (LLMs) are increasingly used to generate labels from radiology reports to enable large-scale AI evaluation. However, label noise from LLMs can introduce bias into performance estimates, especially under varying disease prevalence and model quality. This study quantifies how LLM labeling errors impact downstream diagnostic model evaluation. We developed a simulation framework to assess how LLM label errors affect observed model performance. A synthetic dataset of 10,000 cases was generated across different prevalence levels. LLM sensitivity and specificity were varied independently between 90% and 100%. We simulated diagnostic models with true sensitivity and specificity ranging from 90% to 100%. Observed performance was computed using LLM-generated labels as the reference. We derived analytical performance bounds and ran 5,000 Monte Carlo trials per condition to estimate empirical uncertainty. Observed performance was highly sensitive to LLM label quality, with bias strongly influenced by disease prevalence. In low-prevalence settings, small reductions in LLM specificity led to substantial underestimation of sensitivity. For example, at 10% prevalence, an LLM with 95% specificity yielded an observed sensitivity of ~53% despite a perfect model. In high-prevalence scenarios, reduced LLM sensitivity caused underestimation of model specificity. Monte Carlo simulations consistently revealed downward bias, with observed performance often falling below true values even when within theoretical bounds. LLM-generated labels can introduce systematic, prevalence-dependent bias into model evaluation. Specificity is more critical in low-prevalence tasks, while sensitivity dominates in high-prevalence settings. These findings highlight the importance of prevalence-aware prompt design and error characterization when using LLMs for post-deployment model assessment in clinical AI.

stat.ME↗

arXiv2025

Evaluating Vision Language Models (VLMs) for Radiology: A Comprehensive Analysis

Foundation models, trained on vast amounts of data using self-supervised techniques, have emerged as a promising frontier for advancing artificial intelligence (AI) applications in medicine. This study evaluates three different vision-language foundation models (RAD-DINO, CheXagent, and BiomedCLIP) on their ability to capture fine-grained imaging features for radiology tasks. The models were assessed across classification, segmentation, and regression tasks for pneumothorax and cardiomegaly on chest radiographs. Self-supervised RAD-DINO consistently excelled in segmentation tasks, while text-supervised CheXagent demonstrated superior classification performance. BiomedCLIP showed inconsistent performance across tasks. A custom segmentation model that integrates global and local features substantially improved performance for all foundation models, particularly for challenging pneumothorax segmentation. The findings highlight that pre-training methodology significantly influences model performance on specific downstream tasks. For fine-grained segmentation tasks, models trained without text supervision performed better, while text-supervised models offered advantages in classification and interpretability. These insights provide guidance for selecting foundation models based on specific clinical applications in radiology.

cs.CV↗

arXiv2025

DMind Benchmark: Toward a Holistic Assessment of LLM Capabilities across the Web3 Domain

The Web3 ecosystem, underpinned by cryptographic primitives and decentralized consensus, represents a high-stakes environment where software vulnerabilities and incentive misalignments translate directly into financial loss. As Large Language Models (LLMs) are increasingly integrated into this domain for tasks ranging from smart contract auditing to decentralized finance analytics, ensuring their reliability is paramount. However, general-purpose benchmarks fail to capture the specialized reasoning required for these adversarial and protocol-driven settings. To bridge this gap, we introduce DMind Benchmark, a comprehensive evaluation suite designed to rigorously assess LLM proficiency across the Web3 stack. DMind Benchmark encompasses nine distinct subdomains (spanning infrastructure, smart contracts, token economics, etc.) and combines objective knowledge retrieval with complex open-ended reasoning tasks that emulate real-world operational challenges. We conduct an extensive evaluation of 31 leading proprietary and open-weights models, employing a contamination-aware pipeline and verifying the statistical robustness of our scoring protocol through rigorous cross-judge consistency checks. Our analysis reveals a critical dichotomy: while models demonstrate competence in foundational infrastructure concepts, they exhibit significant vulnerabilities in high-reasoning tasks such as security auditing. Furthermore, we provide a Pareto analysis to guide cost-effective deployment and demonstrate through adversarial experiments that high performance on DMind Benchmark necessitates genuine reasoning rather than superficial memorization. Since its open-source release in April 2025, DMind Benchmark achieved the #1 trending position on Hugging Face for nearly a week and accumulated over 13k downloads by June 2026, establishing itself as a standard for advancing secure and trustworthy AI in Web3.

cs.CR↗

arXiv2025

Subgroup Performance of a Commercial Digital Breast Tomosynthesis Model for Breast Cancer Detection

While research has established the potential of AI models for mammography to improve breast cancer screening outcomes, there have not been any detailed subgroup evaluations performed to assess the strengths and weaknesses of commercial models for digital breast tomosynthesis (DBT) imaging. This study presents a granular evaluation of the Lunit INSIGHT DBT model on a large retrospective cohort of 163,449 screening mammography exams from the Emory Breast Imaging Dataset (EMBED). Model performance was evaluated in a binary context with various negative exam types (162,081 exams) compared against screen detected cancers (1,368 exams) as the positive class. The analysis was stratified across demographic, imaging, and pathologic subgroups to identify potential disparities. The model achieved an overall AUC of 0.91 (95% CI: 0.90-0.92) with a precision of 0.08 (95% CI: 0.08-0.08), and a recall of 0.73 (95% CI: 0.71-0.76). Performance was found to be robust across demographics, but cases with non-invasive cancers (AUC: 0.85, 95% CI: 0.83-0.87), calcifications (AUC: 0.80, 95% CI: 0.78-0.82), and dense breast tissue (AUC: 0.90, 95% CI: 0.88-0.91) were associated with significantly lower performance compared to other groups. These results highlight the need for detailed evaluation of model characteristics and vigilance in considering adoption of new tools for clinical deployment.

eess.IV↗

arXiv2025

Novel AI-Based Quantification of Breast Arterial Calcification to Predict Cardiovascular Risk

Women are underdiagnosed and undertreated for cardiovascular disease. Automatic quantification of breast arterial calcification on screening mammography can identify women at risk for cardiovascular disease and enable earlier treatment and management of disease. In this retrospective study of 116,135 women from two healthcare systems, a transformer-based neural network quantified BAC severity (no BAC, mild, moderate, and severe) on screening mammograms. Outcomes included major adverse cardiovascular events (MACE) and all-cause mortality. BAC severity was independently associated with MACE after adjusting for cardiovascular risk factors, with increasing hazard ratios from mild (HR 1.18-1.22), moderate (HR 1.38-1.47), to severe BAC (HR 2.03-2.22) across datasets (all p<0.001). This association remained significant across all age groups, with even mild BAC indicating increased risk in women under 50. BAC remained an independent predictor when analyzed alongside ASCVD risk scores, showing significant associations with myocardial infarction, stroke, heart failure, and mortality (all p<0.005). Automated BAC quantification enables opportunistic cardiovascular risk assessment during routine mammography without additional radiation or cost. This approach provides value beyond traditional risk factors, particularly in younger women, offering potential for early CVD risk stratification in the millions of women undergoing annual mammography.

eess.IV↗

arXiv2024

I Experienced More than 10 DeFi Scams: On DeFi Users' Perception of Security Breaches and Countermeasures

Decentralized Finance (DeFi) offers a whole new investment experience and has quickly emerged as an enticing alternative to Centralized Finance (CeFi). Rapidly growing market size and active users, however, have also made DeFi a lucrative target for scams and hacks, with 1.95 billion USD lost in 2023. Unfortunately, no prior research thoroughly investigates DeFi users' security risk awareness levels and the adequacy of their risk mitigation strategies. Based on a semi-structured interview study (N = 14) and a follow-up survey (N = 493), this paper investigates DeFi users' security perceptions and commonly adopted practices, and how those affected by previous scams or hacks (DeFi victims) respond and try to recover their losses. Our analysis shows that users often prefer DeFi over CeFi due to their decentralized nature and strong profitability. Despite being aware that DeFi, compared to CeFi, is prone to more severe attacks, users are willing to take those risks to explore new investment opportunities. Worryingly, most victims do not learn from previous experiences; unlike victims studied through traditional systems, DeFi victims tend to find new services, without revising their security practices, to recover their losses quickly. The abundance of various DeFi services and opportunities allows victims to continuously explore new financial opportunities, and this reality seems to cloud their security priorities. Indeed, our results indicate that DeFi users' strong financial motivations outweigh their security concerns - much like those who are addicted to gambling. Our observations about victims' post-incident behaviors suggest that stronger control in the form of industry regulations would be necessary to protect DeFi users from future breaches.

cs.CR↗

arXiv2024

Identifying Multiple Personalities in Large Language Models with External Evaluation

As Large Language Models (LLMs) are integrated with human daily applications rapidly, many societal and ethical concerns are raised regarding the behavior of LLMs. One of the ways to comprehend LLMs' behavior is to analyze their personalities. Many recent studies quantify LLMs' personalities using self-assessment tests that are created for humans. Yet many critiques question the applicability and reliability of these self-assessment tests when applied to LLMs. In this paper, we investigate LLM personalities using an alternate personality measurement method, which we refer to as the external evaluation method, where instead of prompting LLMs with multiple-choice questions in the Likert scale, we evaluate LLMs' personalities by analyzing their responses toward open-ended situational questions using an external machine learning model. We first fine-tuned a Llama2-7B model as the MBTI personality predictor that outperforms the state-of-the-art models as the tool to analyze LLMs' responses. Then, we prompt the LLMs with situational questions and ask them to generate Twitter posts and comments, respectively, in order to assess their personalities when playing two different roles. Using the external personality evaluation method, we identify that the obtained personality types for LLMs are significantly different when generating posts versus comments, whereas humans show a consistent personality profile in these two different situations. This shows that LLMs can exhibit different personalities based on different scenarios, thus highlighting a fundamental difference between personality in LLMs and humans. With our work, we call for a re-evaluation of personality definition and measurement in LLMs.

cs.CL↗