Search

We aim to provide more insights into the applicability to solar coronal seismology of the much-studied discrete leaky modes (DLMs) in classic analyses. Under linear ideal pressureless MHD, we examine two-dimensional (2D) axial fundamental kink motions that arise when localized velocity exciters impact some symmetric slab equilibria. Continuous structuring is allowed for. A 1D initial value problem (IVP) is formulated in conjunction with an eigenvalue problem (EVP) for laterally open systems, with no strict boundary conditions (BCs) at infinity. The IVP is solved by eigenfunction expansion, allowing a clear distinction between the contributions from proper eigenmodes and improper continuum eigenmodes. Example solutions are offered for parameters typical of active region loops. Our solutions show that the system evolves towards long periodicities due to proper eigenmodes (of order the axial Alfven time), whereas the interference of the improper continuum may lead to short periodicities initially (of order the lateral Alfven time). Specializing to the slab axis, we demonstrate that the proper contribution strengthens with the density contrast, but may occasionally be stronger for less steep density profiles. Short periodicities are not guaranteed in the improper contribution, the details of the initial exciter being key. When identifiable, these periodicities tend to agree with the oscillation frequencies expected for DLMs, despite the differences in the BCs between our EVP and classic analyses. The eigenfunction expansion approach enables all qualitative features to be interpreted as the interplay between the initial exciter and some response function, the latter solely determined by the equilibria. Classic theories for DLMs can find seismological applications, with time-dependent studies offering additional ways for constraining initial exciters., Comment: accepted for publication in A&A

This paper introduces a novel framework for unified incremental few-shot object detection (iFSOD) and instance segmentation (iFSIS) using the Transformer architecture. Our goal is to create an optimal solution for situations where only a few examples of novel object classes are available, with no access to training data for base or old classes, while maintaining high performance across both base and novel classes. To achieve this, We extend Mask-DINO into a two-stage incremental learning framework. Stage 1 focuses on optimizing the model using the base dataset, while Stage 2 involves fine-tuning the model on novel classes. Besides, we incorporate a classifier selection strategy that assigns appropriate classifiers to the encoder and decoder according to their distinct functions. Empirical evidence indicates that this approach effectively mitigates the over-fitting on novel classes learning. Furthermore, we implement knowledge distillation to prevent catastrophic forgetting of base classes. Comprehensive evaluations on the COCO and LVIS datasets for both iFSIS and iFSOD tasks demonstrate that our method significantly outperforms state-of-the-art approaches., Comment: 11 pages, 3 figures

With the rapidly increasing capabilities and adoption of code agents for AI-assisted coding, safety concerns, such as generating or executing risky code, have become significant barriers to the real-world deployment of these agents. To provide comprehensive and practical evaluations on the safety of code agents, we propose RedCode, a benchmark for risky code execution and generation: (1) RedCode-Exec provides challenging prompts that could lead to risky code execution, aiming to evaluate code agents' ability to recognize and handle unsafe code. We provide a total of 4,050 risky test cases in Python and Bash tasks with diverse input formats including code snippets and natural text. They covers 25 types of critical vulnerabilities spanning 8 domains (e.g., websites, file systems). We provide Docker environments and design corresponding evaluation metrics to assess their execution results. (2) RedCode-Gen provides 160 prompts with function signatures and docstrings as input to assess whether code agents will follow instructions to generate harmful code or software. Our empirical findings, derived from evaluating three agent frameworks based on 19 LLMs, provide insights into code agents' vulnerabilities. For instance, evaluations on RedCode-Exec show that agents are more likely to reject executing risky operations on the operating system, but are less likely to reject executing technically buggy code, indicating high risks. Risky operations described in natural text lead to a lower rejection rate than those in code format. Additionally, evaluations on RedCode-Gen show that more capable base models and agents with stronger overall coding abilities, such as GPT4, tend to produce more sophisticated and effective harmful software. Our findings highlight the need for stringent safety evaluations for diverse code agents. Our dataset and code are available at https://github.com/AI-secure/RedCode., Comment: Accepted by NeurIPS 2024 Datasets and Benchmarks Track

Multi-instance point cloud registration aims to estimate the pose of all instances of a model point cloud in the whole scene. Existing methods all adopt the strategy of first obtaining the global correspondence and then clustering to obtain the pose of each instance. However, due to the cluttered and occluded objects in the scene, it is difficult to obtain an accurate correspondence between the model point cloud and all instances in the scene. To this end, we propose a simple yet powerful 3D focusing-and-matching network for multi-instance point cloud registration by learning the multiple pair-wise point cloud registration. Specifically, we first present a 3D multi-object focusing module to locate the center of each object and generate object proposals. By using self-attention and cross-attention to associate the model point cloud with structurally similar objects, we can locate potential matching instances by regressing object centers. Then, we propose a 3D dual masking instance matching module to estimate the pose between the model point cloud and each object proposal. It performs instance mask and overlap mask masks to accurately predict the pair-wise correspondence. Extensive experiments on two public benchmarks, Scan2CAD and ROBI, show that our method achieves a new state-of-the-art performance on the multi-instance point cloud registration task. Code is available at https://github.com/zlynpu/3DFMNet., Comment: Accepted to NeurIPS 2024

Differential privacy (DP) is a formal notion that restricts the privacy leakage of an algorithm when running on sensitive data, in which privacy-utility trade-off is one of the central problems in private data analysis. In this work, we investigate the fundamental limits of differential privacy in online learning algorithms and present evidence that separates three types of constraints: no DP, pure DP, and approximate DP. We first describe a hypothesis class that is online learnable under approximate DP but not online learnable under pure DP under the adaptive adversarial setting. This indicates that approximate DP must be adopted when dealing with adaptive adversaries. We then prove that any private online learner must make an infinite number of mistakes for almost all hypothesis classes. This essentially generalizes previous results and shows a strong separation between private and non-private settings since a finite mistake bound is always attainable (as long as the class is online learnable) when there is no privacy requirement.

Machine unlearning has emerged as an effective strategy for forgetting specific information in the training data. However, with the increasing integration of visual data, privacy concerns in Vision Language Models (VLMs) remain underexplored. To address this, we introduce Facial Identity Unlearning Benchmark (FIUBench), a novel VLM unlearning benchmark designed to robustly evaluate the effectiveness of unlearning algorithms under the Right to be Forgotten setting. Specifically, we formulate the VLM unlearning task via constructing the Fictitious Facial Identity VQA dataset and apply a two-stage evaluation pipeline that is designed to precisely control the sources of information and their exposure levels. In terms of evaluation, since VLM supports various forms of ways to ask questions with the same semantic meaning, we also provide robust evaluation metrics including membership inference attacks and carefully designed adversarial privacy attacks to evaluate the performance of algorithms. Through the evaluation of four baseline VLM unlearning algorithms within FIUBench, we find that all methods remain limited in their unlearning performance, with significant trade-offs between model utility and forget quality. Furthermore, our findings also highlight the importance of privacy attacks for robust evaluations. We hope FIUBench will drive progress in developing more effective VLM unlearning algorithms.

Large-scale distributed model training requires simultaneous training on up to thousands of machines. Faulty machine detection is critical when an unexpected fault occurs in a machine. From our experience, a training task can encounter two faults per day on average, possibly leading to a halt for hours. To address the drawbacks of the time-consuming and labor-intensive manual scrutiny, we propose Minder, an automatic faulty machine detector for distributed training tasks. The key idea of Minder is to automatically and efficiently detect faulty distinctive monitoring metric patterns, which could last for a period before the entire training task comes to a halt. Minder has been deployed in our production environment for over one year, monitoring daily distributed training tasks where each involves up to thousands of machines. In our real-world fault detection scenarios, Minder can accurately and efficiently react to faults within 3.6 seconds on average, with a precision of 0.904 and F1-score of 0.893.

Large language models (LLMs) achieve good performance on challenging reasoning benchmarks, yet could also make basic reasoning mistakes. This contrasting behavior is puzzling when it comes to understanding the mechanisms behind LLMs' reasoning capabilities. One hypothesis is that the increasingly high and nearly saturated performance on common reasoning benchmarks could be due to the memorization of similar problems. In this paper, we systematically investigate this hypothesis with a quantitative measurement of memorization in reasoning tasks, using a dynamically generated logical reasoning benchmark based on Knights and Knaves (K&K) puzzles. We found that LLMs could interpolate the training puzzles (achieving near-perfect accuracy) after fine-tuning, yet fail when those puzzles are slightly perturbed, suggesting that the models heavily rely on memorization to solve those training puzzles. On the other hand, we show that while fine-tuning leads to heavy memorization, it also consistently improves generalization performance. In-depth analyses with perturbation tests, cross difficulty-level transferability, probing model internals, and fine-tuning with wrong answers suggest that the LLMs learn to reason on K&K puzzles despite training data memorization. This phenomenon indicates that LLMs exhibit a complex interplay between memorization and genuine reasoning abilities. Finally, our analysis with per-sample memorization score sheds light on how LLMs switch between reasoning and memorization in solving logical puzzles. Our code and data are available at https://memkklogic.github.io.

From the perspective of resource-theoretic approach, this study explores the quantification of imaginary in quantum physics. We propose a well defined measure of imaginarity, the geometric-like measure of imaginarity. Compared with the usual geometric imaginarity measure, this geometric-like measure of imaginarity exhibits smaller decay difference under quantum noisy channels and higher stability. As applications, we show that both the optimal probability of state transformations from a pure state to an arbitrary mixed state via real operations, and the maximal probability of stochastic-approximate state transformations from a pure state to an arbitrary mixed state via real operations with a given fidelity $f$, are given by the geometric-like measure of imaginarity.

Despite their success, unsupervised domain adaptation methods for semantic segmentation primarily focus on adaptation between image domains and do not utilize other abundant visual modalities like depth, infrared and event. This limitation hinders their performance and restricts their application in real-world multimodal scenarios. To address this issue, we propose Modality Adaptation with text-to-image Diffusion Models (MADM) for semantic segmentation task which utilizes text-to-image diffusion models pre-trained on extensive image-text pairs to enhance the model's cross-modality capabilities. Specifically, MADM comprises two key complementary components to tackle major challenges. First, due to the large modality gap, using one modal data to generate pseudo labels for another modality suffers from a significant drop in accuracy. To address this, MADM designs diffusion-based pseudo-label generation which adds latent noise to stabilize pseudo-labels and enhance label accuracy. Second, to overcome the limitations of latent low-resolution features in diffusion models, MADM introduces the label palette and latent regression which converts one-hot encoded labels into the RGB form by palette and regresses them in the latent space, thus ensuring the pre-trained decoder for up-sampling to obtain fine-grained features. Extensive experimental results demonstrate that MADM achieves state-of-the-art adaptation performance across various modality tasks, including images to depth, infrared, and event modalities. We open-source our code and models at https://github.com/XiaRho/MADM., Comment: NeurIPS 2024

Vision Language Models (VLMs) have revolutionized the creation of generalist web agents, empowering them to autonomously complete diverse tasks on real-world websites, thereby boosting human efficiency and productivity. However, despite their remarkable capabilities, the safety and security of these agents against malicious attacks remain critically underexplored, raising significant concerns about their safe deployment. To uncover and exploit such vulnerabilities in web agents, we provide AdvWeb, a novel black-box attack framework designed against web agents. AdvWeb trains an adversarial prompter model that generates and injects adversarial prompts into web pages, misleading web agents into executing targeted adversarial actions such as inappropriate stock purchases or incorrect bank transactions, actions that could lead to severe real-world consequences. With only black-box access to the web agent, we train and optimize the adversarial prompter model using DPO, leveraging both successful and failed attack strings against the target agent. Unlike prior approaches, our adversarial string injection maintains stealth and control: (1) the appearance of the website remains unchanged before and after the attack, making it nearly impossible for users to detect tampering, and (2) attackers can modify specific substrings within the generated adversarial string to seamlessly change the attack objective (e.g., purchasing stocks from a different company), enhancing attack flexibility and efficiency. We conduct extensive evaluations, demonstrating that AdvWeb achieves high success rates in attacking SOTA GPT-4V-based VLM agent across various web tasks. Our findings expose critical vulnerabilities in current LLM/VLM-based agents, emphasizing the urgent need for developing more reliable web agents and effective defenses. Our code and data are available at https://ai-secure.github.io/AdvWeb/ ., Comment: 15 pages

In this paper, we study the allocation of indivisible chores and consider the problem of finding a fair allocation that is approximately efficient. We shift our attention from the multiplicative approximation to the additive one. Our results are twofold, with (1) bounding how the optimal social cost escalates resulting from fairness requirements and (2) presenting the hardness of approximation for the problems of finding fair allocations with the minimum social cost. To quantify the escalation, we introduce cost of fairness (CoF) $\unicode{x2014}$ an alternative to the price of fairness (PoF) $\unicode{x2014}$ to bound the difference (v.s. ratio for PoF) between the optimal social cost with and without fairness constraints in the worst-case instance. We find that CoF is more informative than PoF for chores in the sense that the PoF is infinity regarding all EQX (equitable up to any item), EQ1 (equitable up to one item) and EF1 (envy-free up to one item), while the CoF is $n$ regarding EQX and 1 regarding EQ1 and EF1, where $n$ is the number of agents. For inapproximability, we present a detailed picture of hardness of approximation. We prove that finding the optimal EQX allocation within an additive approximation factor of $n$ is NP-hard for any $n \geq 2$ where $n$ is the number of agents and the cost functions are normalized to 1. For EQ1 and EF1, the problem is NP-hard when the additive factor is a constant and $n \geq 3$. When $n = 2$, we design additive approximation schemes for EQ1 and EF1., Comment: Appears in the 20th Conference on Web and Internet Economics (WINE), 2024

We present ClearSR, a new method that can better take advantage of latent low-resolution image (LR) embeddings for diffusion-based real-world image super-resolution (Real-ISR). Previous Real-ISR models mostly focus on how to activate more generative priors of text-to-image diffusion models to make the output high-resolution (HR) images look better. However, since these methods rely too much on the generative priors, the content of the output images is often inconsistent with the input LR ones. To mitigate the above issue, in this work, we explore using latent LR embeddings to constrain the control signals from ControlNet, and extract LR information at both detail and structure levels. We show that the proper use of latent LR embeddings can produce higher-quality control signals, which enables the super-resolution results to be more consistent with the LR image and leads to clearer visual results. In addition, we also show that latent LR embeddings can be used to control the inference stage, allowing for the improvement of fidelity and generation ability simultaneously. Experiments demonstrate that our model can achieve better performance across multiple metrics on several test sets and generate more consistent SR results with LR images than existing methods. Our code will be made publicly available.

Real-world image super-resolution (Real-ISR) aims at restoring high-quality (HQ) images from low-quality (LQ) inputs corrupted by unknown and complex degradations. In particular, pretrained text-to-image (T2I) diffusion models provide strong generative priors to reconstruct credible and intricate details. However, T2I generation focuses on semantic consistency while Real-ISR emphasizes pixel-level reconstruction, which hinders existing methods from fully exploiting diffusion priors. To address this challenge, we introduce ConsisSR to handle both semantic and pixel-level consistency. Specifically, compared to coarse-grained text prompts, we exploit the more powerful CLIP image embedding and effectively leverage both modalities through our Hybrid Prompt Adapter (HPA) for semantic guidance. Secondly, we introduce Time-aware Latent Augmentation (TALA) to mitigate the inherent gap between T2I generation and Real-ISR consistency requirements. By randomly mixing LQ and HQ latent inputs, our model not only handle timestep-specific diffusion noise but also refine the accumulated latent representations. Last but not least, our GAN-Embedding strategy employs the pretrained Real-ESRGAN model to refine the diffusion start point. This accelerates the inference process to 10 steps while preserving sampling quality, in a training-free manner. Our method demonstrates state-of-the-art performance among both full-scale and accelerated models. The code will be made publicly available.

Perceiving and generating diverse modalities are crucial for AI models to effectively learn from and engage with real-world signals, necessitating reliable evaluations for their development. We identify two major issues in current evaluations: (1) inconsistent standards, shaped by different communities with varying protocols and maturity levels; and (2) significant query, grading, and generalization biases. To address these, we introduce MixEval-X, the first any-to-any, real-world benchmark designed to optimize and standardize evaluations across diverse input and output modalities. We propose multi-modal benchmark mixture and adaptation-rectification pipelines to reconstruct real-world task distributions, ensuring evaluations generalize effectively to real-world use cases. Extensive meta-evaluations show our approach effectively aligns benchmark samples with real-world task distributions. Meanwhile, MixEval-X's model rankings correlate strongly with that of crowd-sourced real-world evaluations (up to 0.98) while being much more efficient. We provide comprehensive leaderboards to rerank existing models and organizations and offer insights to enhance understanding of multi-modal evaluations and inform future research.

Differential privacy (DP) is the de facto privacy standard against privacy leakage attacks, including many recently discovered ones against large language models (LLMs). However, we discovered that LLMs could reconstruct the altered/removed privacy from given DP-sanitized prompts. We propose two attacks (black-box and white-box) based on the accessibility to LLMs and show that LLMs could connect the pair of DP-sanitized text and the corresponding private training data of LLMs by giving sample text pairs as instructions (in the black-box attacks) or fine-tuning data (in the white-box attacks). To illustrate our findings, we conduct comprehensive experiments on modern LLMs (e.g., LLaMA-2, LLaMA-3, ChatGPT-3.5, ChatGPT-4, ChatGPT-4o, Claude-3, Claude-3.5, OPT, GPT-Neo, GPT-J, Gemma-2, and Pythia) using commonly used datasets (such as WikiMIA, Pile-CC, and Pile-Wiki) against both word-level and sentence-level DP. The experimental results show promising recovery rates, e.g., the black-box attacks against the word-level DP over WikiMIA dataset gave 72.18% on LLaMA-2 (70B), 82.39% on LLaMA-3 (70B), 75.35% on Gemma-2, 91.2% on ChatGPT-4o, and 94.01% on Claude-3.5 (Sonnet). More urgently, this study indicates that these well-known LLMs have emerged as a new security risk for existing DP text sanitization approaches in the current environment.

Recent research in causal inference under network interference has explored various experimental designs and estimation techniques to address this issue. However, existing methods, which typically rely on single experiments, often reach a performance bottleneck and face limitations in handling diverse interference structures. In contrast, we propose leveraging multiple experiments to overcome these limitations. In industry, the use of sequential experiments, often known as the ramp-up process, where traffic to the treatment gradually increases, is common due to operational needs like risk management and cost control. Our approach shifts the focus from operational aspects to the statistical advantages of merging data from multiple experiments. By combining data from sequentially conducted experiments, we aim to estimate the global average treatment effect more effectively. In this paper, we begin by analyzing the bias and variance of the linear regression estimator for GATE under general linear network interference. We demonstrate that bias plays a dominant role in the bias-variance tradeoff and highlight the intrinsic bias reduction achieved by merging data from experiments with strictly different treatment proportions. Herein the improvement introduced by merging two steps of experimental data is essential. In addition, we show that merging more steps of experimental data is unnecessary under general linear interference, while it can become beneficial when nonlinear interference occurs. Furthermore, we look into a more advanced estimator based on graph neural networks. Through extensive simulation studies, we show that the regression-based estimator benefits remarkably from training on merged experiment data, achieving outstanding statistical performance.

To alleviate hardware scarcity in training large deep neural networks (DNNs), particularly large language models (LLMs), we present FusionLLM, a decentralized training system designed and implemented for training DNNs using geo-distributed GPUs across different computing clusters or individual devices. Decentralized training faces significant challenges regarding system design and efficiency, including: 1) the need for remote automatic differentiation (RAD), 2) support for flexible model definitions and heterogeneous software, 3) heterogeneous hardware leading to low resource utilization or the straggler problem, and 4) slow network communication. To address these challenges, in the system design, we represent the model as a directed acyclic graph of operators (OP-DAG). Each node in the DAG represents the operator in the DNNs, while the edge represents the data dependency between operators. Based on this design, 1) users are allowed to customize any DNN without caring low-level operator implementation; 2) we enable the task scheduling with the more fine-grained sub-tasks, offering more optimization space; 3) a DAG runtime executor can implement RAD withour requiring the consistent low-level ML framework versions. To enhance system efficiency, we implement a workload estimator and design an OP-Fence scheduler to cluster devices with similar bandwidths together and partition the DAG to increase throughput. Additionally, we propose an AdaTopK compressor to adaptively compress intermediate activations and gradients at the slowest communication links. To evaluate the convergence and efficiency of our system and algorithms, we train ResNet-101 and GPT-2 on three real-world testbeds using 48 GPUs connected with 8 Mbps~10 Gbps networks. Experimental results demonstrate that our system and method can achieve 1.45 - 9.39x speedup compared to baseline methods while ensuring convergence.

Existing works have established multiple benchmarks to highlight the security risks associated with Code GenAI. These risks are primarily reflected in two areas: a model potential to generate insecure code (insecure coding) and its utility in cyberattacks (cyberattack helpfulness). While these benchmarks have made significant strides, there remain opportunities for further improvement. For instance, many current benchmarks tend to focus more on a model ability to provide attack suggestions rather than its capacity to generate executable attacks. Additionally, most benchmarks rely heavily on static evaluation metrics, which may not be as precise as dynamic metrics such as passing test cases. Conversely, expert-verified benchmarks, while offering high-quality data, often operate at a smaller scale. To address these gaps, we develop SecCodePLT, a unified and comprehensive evaluation platform for code GenAIs' risks. For insecure code, we introduce a new methodology for data creation that combines experts with automatic generation. Our methodology ensures the data quality while enabling large-scale generation. We also associate samples with test cases to conduct code-related dynamic evaluation. For cyberattack helpfulness, we set up a real environment and construct samples to prompt a model to generate actual attacks, along with dynamic metrics in our environment. We conduct extensive experiments and show that SecCodePLT outperforms the state-of-the-art (SOTA) benchmark CyberSecEval in security relevance. Furthermore, it better identifies the security risks of SOTA models in insecure coding and cyberattack helpfulness. Finally, we apply SecCodePLT to the SOTA code agent, Cursor, and, for the first time, identify non-trivial security risks in this advanced coding agent.

Cooperation as a self-organized collective behavior plays a significant role in the evolution of ecosystems and human society. Reinforcement learning (RL) offers a new perspective, distinct from imitation learning in evolutionary games, for exploring the mechanisms underlying its emergence. However, most existing studies with the public good game (PGG) employ a self-regarding setup or are on pairwise interaction networks. Players in the real world, however, optimize their policies based not only on their histories but also on the histories of their co-players, and the game is played in a group manner. In the work, we investigate the evolution of cooperation in the PGG under the other-regarding reinforcement learning evolutionary game (OR-RLEG) on hypergraph by combining the Q-learning algorithm and evolutionary game framework, where other players' action history is incorporated and the game is played on hypergraphs. Our results show that as the synergy factor increases, the parameter interval is divided into three distinct regions, the absence of cooperation (AC), medium cooperation (MC), and high cooperation (HC), accompanied by two abrupt transitions in the cooperation level near two transition points, respectively. Interestingly, we identify regular and anti-coordinated chessboard structures in the spatial pattern that positively contribute to the first cooperation transition but adversely affect the second. Furthermore, we provide a theoretical treatment for the first transition with an approximated first transition point and reveal that players with a long-sighted perspective and low exploration rate are more likely to reciprocate kindness with each other, thus facilitating the emergence of cooperation. Our findings contribute to understanding the evolution of human cooperation, where other-regarding information and group interactions are commonplace.

In the realm of high-resolution (HR), fine-grained image segmentation, the primary challenge is balancing broad contextual awareness with the precision required for detailed object delineation, capturing intricate details and the finest edges of objects. Diffusion models, trained on vast datasets comprising billions of image-text pairs, such as SD V2.1, have revolutionized text-to-image synthesis by delivering exceptional quality, fine detail resolution, and strong contextual awareness, making them an attractive solution for high-resolution image segmentation. To this end, we propose DiffDIS, a diffusion-driven segmentation model that taps into the potential of the pre-trained U-Net within diffusion models, specifically designed for high-resolution, fine-grained object segmentation. By leveraging the robust generalization capabilities and rich, versatile image representation prior of the SD models, coupled with a task-specific stable one-step denoising approach, we significantly reduce the inference time while preserving high-fidelity, detailed generation. Additionally, we introduce an auxiliary edge generation task to not only enhance the preservation of fine details of the object boundaries, but reconcile the probabilistic nature of diffusion with the deterministic demands of segmentation. With these refined strategies in place, DiffDIS serves as a rapid object mask generation model, specifically optimized for generating detailed binary maps at high resolutions, while demonstrating impressive accuracy and swift processing. Experiments on the DIS5K dataset demonstrate the superiority of DiffDIS, achieving state-of-the-art results through a streamlined inference process. Our code will be made publicly available., Comment: 13 pages

Power outages have become increasingly frequent, intense, and prolonged in the US due to climate change, aging electrical grids, and rising energy demand. However, largely due to the absence of granular spatiotemporal outage data, we lack data-driven evidence and analytics-based metrics to quantify power system vulnerability. This limitation has hindered the ability to effectively evaluate and address vulnerability to power outages in US communities. Here, we collected ~179 million power outage records at 15-minute intervals across 3022 US contiguous counties (96.15% of the area) from 2014 to 2023. We developed a power system vulnerability assessment framework based on three dimensions (intensity, frequency, and duration) and applied interpretable machine learning models (XGBoost and SHAP) to compute Power System Vulnerability Index (PSVI) at the county level. Our analysis reveals a consistent increase in power system vulnerability over the past decade. We identified 318 counties across 45 states as hotspots for high power system vulnerability, particularly in the West Coast (California and Washington), the East Coast (Florida and the Northeast area), the Great Lakes megalopolis (Chicago-Detroit metropolitan areas), and the Gulf of Mexico (Texas). Heterogeneity analysis indicates that urban counties, counties with interconnected grids, and states with high solar generation exhibit significantly higher vulnerability. Our results highlight the significance of the proposed PSVI for evaluating the vulnerability of communities to power outages. The findings underscore the widespread and pervasive impact of power outages across the country and offer crucial insights to support infrastructure operators, policymakers, and emergency managers in formulating policies and programs aimed at enhancing the resilience of the US power infrastructure.

Graph-based anomaly detection is pivotal in diverse security applications, such as fraud detection in transaction networks and intrusion detection for network traffic. Standard approaches, including Graph Neural Networks (GNNs), often struggle to generalize across shifting data distributions. Meanwhile, real-world domain knowledge is more stable and a common existing component of real-world detection strategies. To explicitly integrate such knowledge into data-driven models such as GCNs, we propose KnowGraph, which integrates domain knowledge with data-driven learning for enhanced graph-based anomaly detection. KnowGraph comprises two principal components: (1) a statistical learning component that utilizes a main model for the overarching detection task, augmented by multiple specialized knowledge models that predict domain-specific semantic entities; (2) a reasoning component that employs probabilistic graphical models to execute logical inferences based on model outputs, encoding domain knowledge through weighted first-order logic formulas. Extensive experiments on these large-scale real-world datasets show that KnowGraph consistently outperforms state-of-the-art baselines in both transductive and inductive settings, achieving substantial gains in average precision when generalizing to completely unseen test graphs. Further ablation studies demonstrate the effectiveness of the proposed reasoning component in improving detection performance, especially under extreme class imbalance. These results highlight the potential of integrating domain knowledge into data-driven models for high-stakes, graph-based security applications., Comment: Accepted to ACM CCS 2024

Recent approaches attempt to adapt powerful interactive segmentation models, such as SAM, to interactive matting and fine-tune the models based on synthetic matting datasets. However, models trained on synthetic data fail to generalize to complex and occlusion scenes. We address this challenge by proposing a new matting dataset based on the COCO dataset, namely COCO-Matting. Specifically, the construction of our COCO-Matting includes accessory fusion and mask-to-matte, which selects real-world complex images from COCO and converts semantic segmentation masks to matting labels. The built COCO-Matting comprises an extensive collection of 38,251 human instance-level alpha mattes in complex natural scenarios. Furthermore, existing SAM-based matting methods extract intermediate features and masks from a frozen SAM and only train a lightweight matting decoder by end-to-end matting losses, which do not fully exploit the potential of the pre-trained SAM. Thus, we propose SEMat which revamps the network architecture and training objectives. For network architecture, the proposed feature-aligned transformer learns to extract fine-grained edge and transparency features. The proposed matte-aligned decoder aims to segment matting-specific objects and convert coarse masks into high-precision mattes. For training objectives, the proposed regularization and trimap loss aim to retain the prior from the pre-trained model and push the matting logits extracted from the mask decoder to contain trimap-based semantic information. Extensive experiments across seven diverse datasets demonstrate the superior performance of our method, proving its efficacy in interactive natural image matting. We open-source our code, models, and dataset at https://github.com/XiaRho/SEMat.

Hot electrons are theoretically predicted to be long-lived in strongly confined quantum dots, which could play vital roles in quantum dot-based optoelectronics; however, existing photoexcitation transient spectroscopy investigations reveal that their lifetime is less than 1 ps in well-passivated quantum dots because of the ultrafast electron-hole Auger-assisted cooling. Therefore, they are generally considered absent in quantum dot optoelectronic devices. Here, by using our newly developed electrically excited transient absorption spectroscopy, we surprisingly observed abundant hot electrons in both II-VI and III-VI compound quantum dot light-emitting diodes at elevated bias (>4 V), of which the lifetimes reach 59 to 371 ns, lengthened by more than 5 orders of magnitude compared with the photoexcited hot electrons. These results experimentally prove the presence of a strong phonon bottleneck effect, refreshing our understanding of the role of hot electrons in quantum dot optoelectronics.

Knowledge graph completion (KGC) tasks aim to infer missing facts in a knowledge graph (KG) for many knowledge-intensive applications. However, existing embedding-based KGC approaches primarily rely on factual triples, potentially leading to outcomes inconsistent with common sense. Besides, generating explicit common sense is often impractical or costly for a KG. To address these challenges, we propose a pluggable common sense-enhanced KGC framework that incorporates both fact and common sense for KGC. This framework is adaptable to different KGs based on their entity concept richness and has the capability to automatically generate explicit or implicit common sense from factual triples. Furthermore, we introduce common sense-guided negative sampling and a coarse-to-fine inference approach for KGs with rich entity concepts. For KGs without concepts, we propose a dual scoring scheme involving a relation-aware concept embedding mechanism. Importantly, our approach can be integrated as a pluggable module for many knowledge graph embedding (KGE) models, facilitating joint common sense and fact-driven training and inference. The experiments illustrate that our framework exhibits good scalability and outperforms existing models across various KGC tasks., Comment: 18 pages, 7 figures, 9 tables

Let $(X,d,\mu)$ be a metric measure space satisfying a doubling property with the upper/lower dimension $Q\ge n>1$, and admitting an $L^2$-Poincar\'e inequality. In this article, we establish the H\"{o}lder continuity and a Liouville-type theorem for the (elliptic-type) Schr\"odinger equation $$\mathbb L u(x,t)=-\partial^2_{t}u(x,t)+\mathcal L u(x,t)+V(x)u(x,t)=0,\quad x\in X,\, t\in\mathbb R, $$ where $\mathcal L$ is a non-negative operator generated by a Dirichlet form on $X$, and the non-negative potential $V$ is a Muckenhoupt weight belonging to the reverse H\"older class ${RH}_q(X)$ for some $q>\max\{Q/2,1\}$. Note that $Q/2$ is critical for the regularity theory of $-\Delta+V$ on $\mathbb{R}^Q$ ($Q\ge3$) by Shen's work in 1995, which hints the critical index of $V$ for the regularity results above on $X\times \mathbb R$ may be $(Q+1)/2$. Our results show that this critical index is in fact $\max\{Q/2,1\}$. Our approach primarily relies on the controllable growth of $V$ and the elliptic theory for the operator $\mathbb L$/$-\partial^2_{t}+\mathcal{L}$ on $X\times \mathbb R$, rather than the analogs for $\mathcal L+V$/$\mathcal{L}$ on $X$, under the critical index setting. As applications, we further obtain some characterizations for solutions to the Schr\"odinger equation $-\partial^2_{t}u+\mathcal L u+Vu=0$ in $X\times \mathbb R_+$ with boundary values in BMO/CMO/Morrey spaces related to $V$, improving previous results to the critical index $q>\max\{Q/2,1\}$., Comment: 39 pages, no figures

In this paper, we propose AutoDAN-Turbo, a black-box jailbreak method that can automatically discover as many jailbreak strategies as possible from scratch, without any human intervention or predefined scopes (e.g., specified candidate strategies), and use them for red-teaming. As a result, AutoDAN-Turbo can significantly outperform baseline methods, achieving a 74.3% higher average attack success rate on public benchmarks. Notably, AutoDAN-Turbo achieves an 88.5 attack success rate on GPT-4-1106-turbo. In addition, AutoDAN-Turbo is a unified framework that can incorporate existing human-designed jailbreak strategies in a plug-and-play manner. By integrating human-designed strategies, AutoDAN-Turbo can even achieve a higher attack success rate of 93.4 on GPT-4-1106-turbo., Comment: Pre-print. Project Page: https://autodans.github.io/AutoDAN-Turbo Code: https://github.com/SaFoLab-WISC/AutoDAN-Turbo

The development of video large multimodal models (LMMs) has been hindered by the difficulty of curating large amounts of high-quality raw data from the web. To address this, we propose an alternative approach by creating a high-quality synthetic dataset specifically for video instruction-following, namely LLaVA-Video-178K. This dataset includes key tasks such as detailed captioning, open-ended question-answering (QA), and multiple-choice QA. By training on this dataset, in combination with existing visual instruction tuning data, we introduce LLaVA-Video, a new video LMM. Our experiments demonstrate that LLaVA-Video achieves strong performance across various video benchmarks, highlighting the effectiveness of our dataset. We plan to release the dataset, its generation pipeline, and the model checkpoints., Comment: Project page: https://llava-vl.github.io/blog/2024-09-30-llava-video/

Recent advancements in large language models (LLMs) have shown promising results in multilingual translation even with limited bilingual supervision. The major challenges are catastrophic forgetting and parameter interference for finetuning LLMs when provided parallel training data. To address these challenges, we propose LANDeRMT, a \textbf{L}anguage-\textbf{A}ware \textbf{N}euron \textbf{De}tecting and \textbf{R}outing framework that selectively finetunes LLMs to \textbf{M}achine \textbf{T}ranslation with diverse translation training data. In LANDeRMT, we evaluate the awareness of neurons to MT tasks and categorize them into language-general and language-specific neurons. This categorization enables selective parameter updates during finetuning, mitigating parameter interference and catastrophic forgetting issues. For the detected neurons, we further propose a conditional awareness-based routing mechanism to dynamically adjust language-general and language-specific capacity within LLMs, guided by translation signals. Experimental results demonstrate that the proposed LANDeRMT is very effective in learning translation knowledge, significantly improving translation quality over various strong baselines for multiple language pairs.

Panoramic X-ray (PX) is a prevalent modality in dentistry practice owing to its wide availability and low cost. However, as a 2D projection of a 3D structure, PX suffers from anatomical information loss and PX diagnosis is limited compared to that with 3D imaging modalities. 2D-to-3D reconstruction methods have been explored for the ability to synthesize the absent 3D anatomical information from 2D PX for use in PX image analysis. However, there are challenges in leveraging such 3D synthesized reconstructions. First, inferring 3D depth from 2D images remains a challenging task with limited accuracy. The second challenge is the joint analysis of 2D PX with its 3D synthesized counterpart, with the aim to maximize the 2D-3D synergy while minimizing the errors arising from the synthesized image. In this study, we propose a new method termed 3DPX - PX image analysis guided by 2D-to-3D reconstruction, to overcome these challenges. 3DPX consists of (i) a novel progressive reconstruction network to improve 2D-to-3D reconstruction and, (ii) a contrastive-guided bidirectional multimodality alignment module for 3D-guided 2D PX classification and segmentation tasks. The reconstruction network progressively reconstructs 3D images with knowledge imposed on the intermediate reconstructions at multiple pyramid levels and incorporates Multilayer Perceptrons to improve semantic understanding. The downstream networks leverage the reconstructed images as 3D anatomical guidance to the PX analysis through feature alignment, which increases the 2D-3D synergy with bidirectional feature projection and decease the impact of potential errors with contrastive guidance. Extensive experiments on two oral datasets involving 464 studies demonstrate that 3DPX outperforms the state-of-the-art methods in various tasks including 2D-to-3D reconstruction, PX classification and lesion segmentation.

We study finite-difference approximations of both Poisson and Poisson-Boltzmann (PB) electrostatic energy functionals for periodic structures constrained by Gauss' law and a class of local algorithms for minimizing the finite-difference discretization of such functionals. The variable of Poisson energy is the vector field of electric displacement and that for the PB energy consists of an electric displacement and ionic concentrations. The displacement is discretized at midpoints of edges of grid boxes while the concentrations are discretize at grid points. The local algorithm is an iteration over all the grid boxes that locally minimizes the energy on each grid box, keeping Gauss' law satisfied. We prove that the energy functionals admit unique minimizers that are solutions to the corresponding Poisson's and charge-conserved PB equation, respectively. Local equilibrium conditions are identified to characterize the finite-difference minimizers of the discretized energy functionals. These conditions are the curl free for the Poisson case and the discrete Boltzmann distributions for the PB case, respectively. Next, we obtain the uniform bound with respect to the grid size h and O(h2)-error estimates in maximum norm for the finite-difference minimizers. The local algorithms are detailed, and a new local algorithm with shift is proposed to treat the general case of a variable coefficient for the Poisson energy. We prove the convergence of all these local algorithms, using the characterization of the finite-difference minimizers. Finally, we present numerical tests to demonstrate the results of our analysis.

We present a new approach for investigating the Markovian to non-Markovian transition in quantum aggregates strongly coupled to a vibrational bath through the analysis of linear absorption spectra. Utilizing hierarchical algebraic equations in the frequency domain, we elucidate how these spectra can effectively reveal transitions between Markovian and non-Markovian regimes, driven by the complex interplay of dissipation, aggregate-bath coupling, and intra-aggregate dipole-dipole interactions. Our results demonstrate that reduced dissipation induces spectral peak splitting, signaling the emergence of bath-induced non-Markovian effects. The spectral peak splitting can also be driven by enhanced dipole-dipole interactions, although the underlying mechanism differs from that of dissipation-induced splitting. Additionally, with an increase in aggregate-bath coupling strength, initially symmetric or asymmetric peaks with varying spectral amplitudes may merge under weak dipole-dipole interactions, whereas strong dipole-dipole interactions are more likely to cause peak splitting. Moreover, we find that spectral features serve as highly sensitive indicators for distinguishing the geometric structures of aggregates, while also unveiling the critical role geometry plays in shaping non-Markovian behavior. This study not only deepens our understanding of the Markovian to non-Markovian transition but also provides a robust framework for optimizing and controlling quantum systems., Comment: 13 pages, 9 figures

Generalist web agents have demonstrated remarkable potential in autonomously completing a wide range of tasks on real websites, significantly boosting human productivity. However, web tasks, such as booking flights, usually involve users' PII, which may be exposed to potential privacy risks if web agents accidentally interact with compromised websites, a scenario that remains largely unexplored in the literature. In this work, we narrow this gap by conducting the first study on the privacy risks of generalist web agents in adversarial environments. First, we present a realistic threat model for attacks on the website, where we consider two adversarial targets: stealing users' specific PII or the entire user request. Then, we propose a novel attack method, termed Environmental Injection Attack (EIA). EIA injects malicious content designed to adapt well to environments where the agents operate and our work instantiates EIA specifically for privacy scenarios in web environments. We collect 177 action steps that involve diverse PII categories on realistic websites from the Mind2Web, and conduct experiments using one of the most capable generalist web agent frameworks to date. The results demonstrate that EIA achieves up to 70% ASR in stealing specific PII and 16% ASR for full user request. Additionally, by accessing the stealthiness and experimenting with a defensive system prompt, we indicate that EIA is hard to detect and mitigate. Notably, attacks that are not well adapted for a webpage can be detected via human inspection, leading to our discussion about the trade-off between security and autonomy. However, extra attackers' efforts can make EIA seamlessly adapted, rendering such supervision ineffective. Thus, we further discuss the defenses at the pre- and post-deployment stages of the websites without relying on human supervision and call for more advanced defense strategies., Comment: 29 pages

The drive to control tokamaks, a prominent technology in nuclear fusion, is essential due to its potential to provide a virtually unlimited source of clean energy. Reinforcement learning (RL) promises improved flexibility to manage the intricate and non-linear dynamics of the plasma encapsulated in a tokamak. However, RL typically requires substantial interaction with a simulator capable of accurately evolving the high-dimensional plasma state. Compared to first-principle-based simulators, whose intense computations lead to sluggish RL training, we devise an effective method to acquire a fully data-driven simulator, by mitigating the arising compounding error issue due to the underlying autoregressive nature. With high accuracy and appealing extrapolation capability, this high-fidelity dynamics model subsequently enables the rapid training of a qualified RL agent to directly generate engineering-reasonable magnetic coil commands, aiming at the desired long-term targets of plasma current and last closed flux surface. Together with a surrogate magnetic equilibrium reconstruction model EFITNN, the RL agent successfully maintains a $100$-ms, $1$ kHz trajectory control with accurate waveform tracking on the HL-3 tokamak. Furthermore, it also demonstrates the feasibility of zero-shot adaptation to changed triangularity targets, confirming the robustness of the developed data-driven dynamics model. Our work underscores the advantage of fully data-driven dynamics models in yielding RL-based trajectory control policies at a sufficiently fast pace, an anticipated engineering requirement in daily discharge practices for the upcoming ITER device.

Significant events such as volcanic eruptions can have global and long lasting impacts on climate. These global impacts, however, are not uniform across space and time. Understanding how the Mt. Pinatubo eruption affects global and regional climate is of great interest for predicting impact on climate due to similar events. We propose a Bayesian framework to simultaneously detect and estimate spatially-varying temporal changepoints for regional climate impacts. Our approach takes into account the diffusing nature of the changes caused by the volcanic eruption and leverages spatial correlation. We illustrate our method on simulated datasets and compare it with an existing changepoint detection method. Finally, we apply our method on monthly stratospheric aerosol optical depth and surface temperature data from 1985 to 1995 to detect and estimate changepoints following the 1991 Mt. Pinatubo eruption.

Multimodal Large Language Models (MLLMs) are evaluated on various benchmarks, such as image captioning, visual question answering, and reasoning. However, many of these benchmarks include overly simple or uninformative samples, complicating the effective distinction of different MLLMs' performance. Furthermore, evaluating models across numerous benchmarks incurs a significant computational burden. To address these issues, we propose LIME (Less Is More for MLLM Evaluation), a refined and efficient benchmark curated through a semi-automated pipeline. This pipeline filters out uninformative samples and eliminates answer leakage by focusing on tasks that necessitate image-based understanding. Our experiments indicate that LIME reduces the number of samples by 76% and evaluation time by 77%, while also providing a more effective means of distinguishing the capabilities of different models. Notably, we find that traditional automatic metrics, such as CIDEr, are inadequate for assessing MLLMs' captioning performance; excluding the caption task score yields a more accurate reflection of overall model performance. All code and data are available at https://github.com/kangreen0210/LIME.

The development of Large Language Models (LLMs) has revolutionized Q&A across various industries, including the database domain. However, there is still a lack of a comprehensive benchmark to evaluate the capabilities of different LLMs and their modular components in database Q&A. To this end, we introduce DQA, the first comprehensive database Q&A benchmark. DQA features an innovative LLM-based method for automating the generation, cleaning, and rewriting of database Q&A, resulting in over 240,000 Q&A pairs in English and Chinese. These Q&A pairs cover nearly all aspects of database knowledge, including database manuals, database blogs, and database tools. This inclusion allows for additional assessment of LLMs' Retrieval-Augmented Generation (RAG) and Tool Invocation Generation (TIG) capabilities in the database Q&A task. Furthermore, we propose a comprehensive LLM-based database Q&A testbed on DQA. This testbed is highly modular and scalable, with both basic and advanced components like Question Classification Routing (QCR), RAG, TIG, and Prompt Template Engineering (PTE). Besides, DQA provides a complete evaluation pipeline, featuring diverse metrics and a standardized evaluation process to ensure comprehensiveness, accuracy, and fairness. We use DQA to evaluate the database Q&A capabilities under the proposed testbed comprehensively. The evaluation reveals findings like (i) the strengths and limitations of nine different LLM-based Q&A bots and (ii) the performance impact and potential improvements of various service components (e.g., QCR, RAG, TIG). We hope our benchmark and findings will better guide the future development of LLM-based database Q&A research., Comment: 12 pages

We study envy-free up to any item (EFX) allocations on graphs where vertices and edges represent agents and items respectively. An agent is only interested in items that are incident to her and all other items have zero marginal values to her. Christodoulou et al. [EC, 2023] first proposed this setting and studied the case of goods. We extend this setting to the case of mixed manna where an item may be liked or disliked by its endpoint agents. In our problem, an agent has an arbitrary valuation over her incident items such that the items she likes have non-negative marginal values to her and those she dislikes have non-positive marginal values. We provide a complete study of the four notions of EFX for mixed manna in the literature, which differ by whether the removed item can have zero marginal value. We prove that an allocation that satisfies the notion of EFX where the virtually-removed item could always have zero marginal value may not exist and determining its existence is NP-complete, while one that satisfies any of the other three notions always exists and can be computed in polynomial time. We also prove that an orientation (i.e., a special allocation where each edge must be allocated to one of its endpoint agents) that satisfies any of the four notions may not exist, and determining its existence is NP-complete., Comment: Accepted in IJCAI 2024

Despite significant anecdotal evidence regarding the vulnerability of the U.S. power infrastructure, there is a dearth of longitudinal and nation-level characterization of the spatial and temporal patterns in the frequency and extent of power outages. A data-driven national-level characterization of power outage vulnerability is particularly essential for understanding the urgency and formulating policies to promote the resilience of power infrastructure systems. Recognizing this, we retrieved 179,053,397 county-level power outage records with a 15-minute interval across 3,022 US counties during 2014-2023 to capture power outage characteristics. We focus on three dimensions--power outage intensity, frequency, and duration--and develop multiple metrics to quantify each dimension of power outage vulnerability. The results show that in the past ten years, the vulnerability of U.S. power system has consistently been increasing. Counties experienced an average of 999.4 outages over the decade, affecting an average of more than 540,000 customers per county, with disruptions occurring approximately every week. Coastal areas, particularly in California, Florida and New Jersey, faced more frequent and prolonged outages, while inland regions showed higher outage rates. A concerning increase in outage frequency and intensity was noted, especially after 2017, with a sharp rise in prolonged outages since 2019. The research also found positive association between social vulnerability and outage metrics, with the association becoming stronger over the years under study. Areas with higher social vulnerability experienced more severe and frequent outages, exacerbating challenges in these regions. These findings reveal the much-needed empirical evidence for stakeholders to inform policy formulation and program development for enhancing the resilience of the U.S. power infrastructure.

Gradient-based meta-learning algorithms have gained popularity for their ability to train models on new tasks using limited data. Empirical observations indicate that such algorithms are able to learn a shared representation across tasks, which is regarded as a key factor in their success. However, the in-depth theoretical understanding of the learning dynamics and the origin of the shared representation remains underdeveloped. In this work, we investigate the meta-learning dynamics of the non-linear two-layer neural networks trained on streaming tasks in the teach-student scenario. Through the lens of statistical physics analysis, we characterize the macroscopic behavior of the meta-training processes, the formation of the shared representation, and the generalization ability of the model on new tasks. The analysis also points to the importance of the choice of certain hyper-parameters of the learning algorithms.

Principal component analysis (PCA) is a classical feature extraction method, but it may be adversely affected by outliers, resulting in inaccurate learning of the projection matrix. This paper proposes a robust method to estimate both the data mean and the PCA projection matrix by learning discriminant sample weights from data containing outliers. Each sample in the dataset is assigned a weight, and the proposed algorithm iteratively learns the weights, the mean, and the projection matrix, respectively. Specifically, when the mean and the projection matrix are available, via fine-grained analysis of outliers, a weight for each sample is learned hierarchically so that outliers have small weights while normal samples have large weights. With the learned weights available, a weighted optimization problem is solved to estimate both the data mean and the projection matrix. Because the learned weights discriminate outliers from normal samples, the adverse influence of outliers is mitigated due to the corresponding small weights. Experiments on toy data, UCI dataset, and face dataset demonstrate the effectiveness of the proposed method in estimating the mean and the projection matrix from the data containing outliers.

Large Language Models (LLMs) have become integral to numerous domains, significantly advancing applications in data management, mining, and analysis. Their profound capabilities in processing and interpreting complex language data, however, bring to light pressing concerns regarding data privacy, especially the risk of unintentional training data leakage. Despite the critical nature of this issue, there has been no existing literature to offer a comprehensive assessment of data privacy risks in LLMs. Addressing this gap, our paper introduces LLM-PBE, a toolkit crafted specifically for the systematic evaluation of data privacy risks in LLMs. LLM-PBE is designed to analyze privacy across the entire lifecycle of LLMs, incorporating diverse attack and defense strategies, and handling various data types and metrics. Through detailed experimentation with multiple LLMs, LLM-PBE facilitates an in-depth exploration of data privacy concerns, shedding light on influential factors such as model size, data characteristics, and evolving temporal dimensions. This study not only enriches the understanding of privacy issues in LLMs but also serves as a vital resource for future research in the field. Aimed at enhancing the breadth of knowledge in this area, the findings, resources, and our full technical report are made available at https://llm-pbe.github.io/, providing an open platform for academic and practical advancements in LLM privacy assessment.

Large Language Models (LLMs) have revolutionized the landscape of machine learning, yet current benchmarks often fall short in capturing the diverse behavior of these models in real-world applications. A benchmark's usefulness is determined by its ability to clearly differentiate between models of varying capabilities (separability) and closely align with human preferences. Existing frameworks like Alpaca-Eval 2.0 LC \cite{dubois2024lengthcontrolledalpacaevalsimpleway} and Arena-Hard v0.1 \cite{li2024crowdsourced} are limited by their focus on general-purpose queries and lack of diversity across domains such as law, medicine, and multilingual contexts. In this paper, we address these limitations by introducing a novel data pipeline that curates diverse, domain-specific evaluation sets tailored for LLM-as-a-Judge frameworks. Our approach leverages a combination of manual curation, semi-supervised learning to generate clusters, and stratified sampling to ensure balanced representation across a wide range of domains and languages. The resulting evaluation set, which includes 1573 samples across 14 categories, demonstrates high separability (84\%) across ten top-ranked models, and agreement (84\%) with Chatbot Arena and (0.915) Spearman correlation. The agreement values are 9\% better than Arena Hard and 20\% better than AlpacaEval 2.0 LC, while the Spearman coefficient is 0.7 more than the next best benchmark, showcasing a significant improvement in the usefulness of the benchmark. We further provide an open-source evaluation tool that enables fine-grained analysis of model performance across user-defined categories, offering valuable insights for practitioners. This work contributes to the ongoing effort to enhance the transparency, diversity, and effectiveness of LLM evaluation methodologies., Comment: 14 pages, 8 figures, Under review