Large language models (LLMs) have been widely deployed as autonomous agents capable of following user instructions and making decisions in real-world applications. Previous studies have made notable progress in benchmarking the instruction following capabilities of LLMs in general domains, with a primary focus on their inherent commonsense knowledge. Recently, LLMs have been increasingly deployed as domain-oriented agents, which rely on domain-oriented guidelines that may conflict with their commonsense knowledge. These guidelines exhibit two key characteristics: they consist of a wide range of domain-oriented rules and are subject to frequent updates. Despite these challenges, the absence of comprehensive benchmarks for evaluating the domain-oriented guideline following capabilities of LLMs presents a significant obstacle to their effective assessment and further development. In this paper, we introduce GuideBench, a comprehensive benchmark designed to evaluate guideline following performance of LLMs. GuideBench evaluates LLMs on three critical aspects: (i) adherence to diverse rules, (ii) robustness to rule updates, and (iii) alignment with human preferences. Experimental results on a range of LLMs indicate substantial opportunities for improving their ability to follow domain-oriented guidelines.
The sparse Mixture-of-Experts (MoE) architecture is increasingly favored for scaling Large Language Models (LLMs) efficiently, but it depends on heterogeneous compute and memory resources. These factors jointly affect system Cost, Accuracy, and Performance (CAP), making trade-offs inevitable. Existing benchmarks often fail to capture these trade-offs accurately, complicating practical deployment decisions. To address this, we introduce MoE-CAP, a benchmark specifically designed for MoE systems. Our analysis reveals that achieving an optimal balance across CAP is difficult with current hardware; MoE systems typically optimize two of the three dimensions at the expense of the third-a dynamic we term the MoE-CAP trade-off. To visualize this, we propose the CAP Radar Diagram. We further introduce sparsity-aware performance metrics-Sparse Memory Bandwidth Utilization (S-MBU) and Sparse Model FLOPS Utilization (S-MFU)-to enable accurate performance benchmarking of MoE systems across diverse hardware platforms and deployment scenarios.
The task of Critical Questions Generation (CQs-Gen) aims to foster critical thinking by enabling systems to generate questions that expose underlying assumptions and challenge the validity of argumentative reasoning structures. Despite growing interest in this area, progress has been hindered by the lack of suitable datasets and automatic evaluation standards. This paper presents a comprehensive approach to support the development and benchmarking of systems for this task. We construct the first large-scale dataset including $~$5K manually annotated questions. We also investigate automatic evaluation methods and propose a reference-based technique using large language models (LLMs) as the strategy that best correlates with human judgments. Our zero-shot evaluation of 11 LLMs establishes a strong baseline while showcasing the difficulty of the task. Data and code plus a public leaderboard are provided to encourage further research not only in terms of model performance, but also to explore the practical benefits of CQs-Gen for both automated reasoning and human critical thinking.
Recent advances in large language models (LLMs) have significantly improved text-to-speech (TTS) systems, enhancing control over speech style, naturalness, and emotional expression, which brings TTS Systems closer to human-level performance. Although the Mean Opinion Score (MOS) remains the standard for TTS System evaluation, it suffers from subjectivity, environmental inconsistencies, and limited interpretability. Existing evaluation datasets also lack a multi-dimensional design, often neglecting factors such as speaking styles, context diversity, and trap utterances, which is particularly evident in Chinese TTS evaluation. To address these challenges, we introduce the Audio Turing Test (ATT), a multi-dimensional Chinese corpus dataset ATT-Corpus paired with a simple, Turing-Test-inspired evaluation protocol. Instead of relying on complex MOS scales or direct model comparisons, ATT asks evaluators to judge whether a voice sounds human. This simplification reduces rating bias and improves evaluation robustness. To further support rapid model development, we also finetune Qwen2-Audio-Instruct with human judgment data as Auto-ATT for automatic evaluation. Experimental results show that ATT effectively differentiates models across specific capability dimensions using its multi-dimensional design. Auto-ATT also demonstrates strong alignment with human evaluations, confirming its value as a fast and reliable assessment tool. The white-box ATT-Corpus and Auto-ATT can be found in ATT Hugging Face Collection (https://huggingface.co/collections/meituan/audio-turing-test-682446320368164faeaf38a4).
When knowledge graphs (KGs) are automatically extracted from text, are they accurate enough for downstream analysis? Unfortunately, current annotated datasets can not be used to evaluate this question, since their KGs are highly disconnected, too small, or overly complex. To address this gap, we introduce AffilKG (https://doi.org/10.5281/zenodo.15427977), which is a collection of six datasets that are the first to pair complete book scans with large, labeled knowledge graphs. Each dataset features affiliation graphs, which are simple KGs that capture Member relationships between Person and Organization entities -- useful in studies of migration, community interactions, and other social phenomena. In addition, three datasets include expanded KGs with a wider variety of relation types. Our preliminary experiments demonstrate significant variability in model performance across datasets, underscoring AffilKG's ability to enable two critical advances: (1) benchmarking how extraction errors propagate to graph-level analyses (e.g., community structure), and (2) validating KG extraction methods for real-world social science research.
Optimizing Large Language Model (LLM) performance requires well-crafted prompts, but manual prompt engineering is labor-intensive and often ineffective. Automated prompt optimization techniques address this challenge but the majority of them rely on randomly selected evaluation subsets, which fail to represent the full dataset, leading to unreliable evaluations and suboptimal prompts. Existing coreset selection methods, designed for LLM benchmarking, are unsuitable for prompt optimization due to challenges in clustering similar samples, high data collection costs, and the unavailability of performance data for new or private datasets. To overcome these issues, we propose IPOMP, an Iterative evaluation data selection for effective Prompt Optimization using real-time Model Performance. IPOMP is a two-stage approach that selects representative and diverse samples using semantic clustering and boundary analysis, followed by iterative refinement with real-time model performance data to replace redundant samples. Evaluations on the BIG-bench dataset show that IPOMP improves effectiveness by 1.6% to 5.3% and stability by at least 57% compared with SOTA baselines, with minimal computational overhead below 1%. Furthermore, the results demonstrate that our real-time performance-guided refinement approach can be universally applied to enhance existing coreset selection methods.
Deep generative models have shown significant promise in generating valid 3D molecular structures, with the GEOM-Drugs dataset serving as a key benchmark. However, current evaluation protocols suffer from critical flaws, including incorrect valency definitions, bugs in bond order calculations, and reliance on force fields inconsistent with the reference data. In this work, we revisit GEOM-Drugs and propose a corrected evaluation framework: we identify and fix issues in data preprocessing, construct chemically accurate valency tables, and introduce a GFN2-xTB-based geometry and energy benchmark. We retrain and re-evaluate several leading models under this framework, providing updated performance metrics and practical recommendations for future benchmarking. Our results underscore the need for chemically rigorous evaluation practices in 3D molecular generation. Our recommended evaluation methods and GEOM-Drugs processing scripts are available at https://github.com/isayevlab/geom-drugs-3dgen-evaluation.
The intrinsic complexity of human biology presents ongoing challenges to scientific understanding. Researchers collaborate across disciplines to expand our knowledge of the biological interactions that define human life. AI methodologies have emerged as powerful tools across scientific domains, particularly in computational biology, where graph data structures effectively model biological entities such as protein-protein interaction (PPI) networks and gene functional networks. Those networks are used as datasets for paramount network medicine tasks, such as gene-disease association prediction, drug repurposing, and polypharmacy side effect studies. Reliable predictions from machine learning models require high-quality foundational data. In this work, we present a comprehensive multi-purpose biological knowledge graph constructed by integrating and refining multiple publicly available datasets. Building upon the Drug Repurposing Knowledge Graph (DRKG), we define a pipeline tasked with a) cleaning inconsistencies and redundancies present in DRKG, b) coalescing information from the main available public data sources, and c) enriching the graph nodes with expressive feature vectors such as molecular fingerprints and gene ontologies. Biologically and chemically relevant features improve the capacity of machine learning models to generate accurate and well-structured embedding spaces. The resulting resource represents a coherent and reliable biological knowledge graph that serves as a state-of-the-art platform to advance research in computational biology and precision medicine. Moreover, it offers the opportunity to benchmark graph-based machine learning and network medicine models on relevant tasks. We demonstrate the effectiveness of the proposed dataset by benchmarking it against the task of drug repurposing, PPI prediction, and side-effect prediction, modeled as link prediction problems.
We present GNN-Suite, a robust modular framework for constructing and benchmarking Graph Neural Network (GNN) architectures in computational biology. GNN-Suite standardises experimentation and reproducibility using the Nextflow workflow to evaluate GNN performance. We demonstrate its utility in identifying cancer-driver genes by constructing molecular networks from protein-protein interaction (PPI) data from STRING and BioGRID and annotating nodes with features from the PCAWG, PID, and COSMIC-CGC repositories. Our design enables fair comparisons among diverse GNN architectures including GAT, GAT3H, GCN, GCN2, GIN, GTN, HGCN, PHGCN, and GraphSAGE and a baseline Logistic Regression (LR) model. All GNNs were configured as standardised two-layer models and trained with uniform hyperparameters (dropout = 0.2; Adam optimiser with learning rate = 0.01; and an adjusted binary cross-entropy loss to address class imbalance) over an 80/20 train-test split for 300 epochs. Each model was evaluated over 10 independent runs with different random seeds to yield statistically robust performance metrics, with balanced accuracy (BACC) as the primary measure. Notably, GCN2 achieved the highest BACC (0.807 +/- 0.035) on a STRING-based network, although all GNN types outperformed the LR baseline, highlighting the advantage of network-based learning over feature-only approaches. Our results show that a common framework for implementing and evaluating GNN architectures aids in identifying not only the best model but also the most effective means of incorporating complementary data. By making GNN-Suite publicly available, we aim to foster reproducible research and promote improved benchmarking standards in computational biology. Future work will explore additional omics datasets and further refine network architectures to enhance predictive accuracy and interpretability in biomedical applications.
HumaniBench는 대규모 다중모달 모델(LMM)의 인간 중심 평가를 위한 종합 벤치마크로, 공정성, 윤리성, 공감능력, 포용성 등 7가지 인간 중심 AI(HCAI) 원칙을 포함한 32,000개의 실제 이미지-질문 쌍으로 구성되어 있습니다. GPT4o 기반의 확장 가능한 주석 파이프라인과 전문가 검증을 통해 데이터가 구축되었으며, 시각적 질문응답(VQA), 다국어 QA, 시각적 근거 찾기, 공감 캡셔닝, 강인성 테스트 등 7가지 다양한 과제를 포함합니다. 15개의 최신 LMM을 평가한 결과, 상용 모델이 전반적으로 우수하지만 강인성과 시각적 근거 찾기에서 약점을 보였고, 일부 오픈소스 모델은 정확성과 인간 중심 원칙 준수 간 균형에 어려움을 겪는 것으로 나타났습니다. HumaniBench는 HCAI 원칙에 특화된 최초의 벤치마크로, 모델의 사회적 책임과 정확성 간 정렬 문제를 진단하고 개선 방향을 제시하는 데 기여합니다.
Spiking Transformers have recently emerged as promising architectures for combining the efficiency of spiking neural networks with the representational power of self-attention. However, the lack of standardized implementations, evaluation pipelines, and consistent design choices has hindered fair comparison and principled analysis. In this paper, we introduce \textbf{STEP}, a unified benchmark framework for Spiking Transformers that supports a wide range of tasks, including classification, segmentation, and detection across static, event-based, and sequential datasets. STEP provides modular support for diverse components such as spiking neurons, input encodings, surrogate gradients, and multiple backends (e.g., SpikingJelly, BrainCog). Using STEP, we reproduce and evaluate several representative models, and conduct systematic ablation studies on attention design, neuron types, encoding schemes, and temporal modeling capabilities. We also propose a unified analytical model for energy estimation, accounting for spike sparsity, bitwidth, and memory access, and show that quantized ANNs may offer comparable or better energy efficiency. Our results suggest that current Spiking Transformers rely heavily on convolutional frontends and lack strong temporal modeling, underscoring the need for spike-native architectural innovations. The full code is available at: https://github.com/Fancyssc/STEP
Robust machine learning depends on clean data, yet current image data cleaning benchmarks rely on synthetic noise or narrow human studies, limiting comparison and real-world relevance. We introduce CleanPatrick, the first large-scale benchmark for data cleaning in the image domain, built upon the publicly available Fitzpatrick17k dermatology dataset. We collect 496,377 binary annotations from 933 medical crowd workers, identify off-topic samples (4%), near-duplicates (21%), and label errors (22%), and employ an aggregation model inspired by item-response theory followed by expert review to derive high-quality ground truth. CleanPatrick formalizes issue detection as a ranking task and adopts typical ranking metrics mirroring real audit workflows. Benchmarking classical anomaly detectors, perceptual hashing, SSIM, Confident Learning, NoiseRank, and SelfClean, we find that, on CleanPatrick, self-supervised representations excel at near-duplicate detection, classical methods achieve competitive off-topic detection under constrained review budgets, and label-error detection remains an open challenge for fine-grained medical classification. By releasing both the dataset and the evaluation framework, CleanPatrick enables a systematic comparison of image-cleaning strategies and paves the way for more reliable data-centric artificial intelligence.
The practical deployment of Visual Anomaly Detection (VAD) systems is hindered by their sensitivity to real-world imaging variations, particularly the complex interplay between viewpoint and illumination which drastically alters defect visibility. Current benchmarks largely overlook this critical challenge. We introduce Multi-View Multi-Illumination Anomaly Detection (M2AD), a new large-scale benchmark comprising 119,880 high-resolution images designed explicitly to probe VAD robustness under such interacting conditions. By systematically capturing 999 specimens across 10 categories using 12 synchronized views and 10 illumination settings (120 configurations total), M2AD enables rigorous evaluation. We establish two evaluation protocols: M2AD-Synergy tests the ability to fuse information across diverse configurations, and M2AD-Invariant measures single-image robustness against realistic view-illumination effects. Our extensive benchmarking shows that state-of-the-art VAD methods struggle significantly on M2AD, demonstrating the profound challenge posed by view-illumination interplay. This benchmark serves as an essential tool for developing and validating VAD methods capable of overcoming real-world complexities. Our full dataset and test suite will be released at https://hustcyq.github.io/M2AD to facilitate the field.
The rapid extension of context windows in large vision-language models has given rise to long-context vision-language models (LCVLMs), which are capable of handling hundreds of images with interleaved text tokens in a single forward pass. In this work, we introduce MMLongBench, the first benchmark covering a diverse set of long-context vision-language tasks, to evaluate LCVLMs effectively and thoroughly. MMLongBench is composed of 13,331 examples spanning five different categories of downstream tasks, such as Visual RAG and Many-Shot ICL. It also provides broad coverage of image types, including various natural and synthetic images. To assess the robustness of the models to different input lengths, all examples are delivered at five standardized input lengths (8K-128K tokens) via a cross-modal tokenization scheme that combines vision patches and text tokens. Through a thorough benchmarking of 46 closed-source and open-source LCVLMs, we provide a comprehensive analysis of the current models' vision-language long-context ability. Our results show that: i) performance on a single task is a weak proxy for overall long-context capability; ii) both closed-source and open-source models face challenges in long-context vision-language tasks, indicating substantial room for future improvement; iii) models with stronger reasoning ability tend to exhibit better long-context performance. By offering wide task coverage, various image types, and rigorous length control, MMLongBench provides the missing foundation for diagnosing and advancing the next generation of LCVLMs.
Brain-Computer Interfaces enable direct communication between the brain and external systems, with functional Near-Infrared Spectroscopy emerging as a portable and non-invasive method for capturing cerebral hemodynamics. This study investigates the classification of rest and task states during a realistic, interactive tennis simulation using fNIRS signals and a range of machine learning approaches. We benchmarked traditional classifiers based on engineered features, Long Short-Term Memory networks on raw time-series data, and Convolutional Neural Networks applied to Gramian Angular Field-transformed images. Ensemble models like Extra Trees and Gradient Boosting achieved accuracies above 97 percent, while the ResNet-based CNN reached 95.0 percent accuracy with a near-perfect AUC of 99.2 percent, outperforming both LSTM and EfficientNet architectures. A novel data augmentation strategy was employed to equalize trial durations while preserving physiological integrity. Feature importance analyses revealed that both oxygenated and deoxygenated hemoglobin signals, particularly slope and RMS metrics, were key contributors to classification performance. These findings demonstrate the strong potential of fNIRS-based BCIs for deployment in dynamic, real-world environments and underscore the advantages of deep learning models in decoding complex neural signals.
In this paper, we introduce S-MedQA, an English medical question-answering (QA) dataset for benchmarking large language models in fine-grained clinical specialties. We use S-MedQA to check the applicability of a popular hypothesis related to knowledge injection in the knowledge-intense scenario of medical QA, and show that: 1) training on data from a speciality does not necessarily lead to best performance on that specialty and 2) regardless of the specialty fine-tuned on, token probabilities of clinically relevant terms for all specialties increase consistently. Thus, we believe improvement gains come mostly from domain shifting (e.g., general to medical) rather than knowledge injection and suggest rethinking the role of fine-tuning data in the medical domain. We release S-MedQA and all code needed to reproduce all our experiments to the research community.
Depth estimation and scene segmentation are two important tasks in intelligent transportation systems. A joint modeling of these two tasks will reduce the requirement for both the storage and training efforts. This work explores how the multi-task distillation could be used to improve such unified modeling. While existing solutions transfer multiple teachers' knowledge in a static way, we propose a self-adaptive distillation method that can dynamically adjust the knowledge amount from each teacher according to the student's current learning ability. Furthermore, as multiple teachers exist, the student's gradient update direction in the distillation is more prone to be erroneous where knowledge forgetting may occur. To avoid this, we propose a knowledge trajectory to record the most essential information that a model has learnt in the past, based on which a trajectory-based distillation loss is designed to guide the student to follow the learning curve similarly in a cost-effective way. We evaluate our method on multiple benchmarking datasets including Cityscapes and NYU-v2. Compared to the state-of-the-art solutions, our method achieves a clearly improvement. The code is provided in the supplementary materials.
Despite significant advancements in Deep Reinforcement Learning (DRL) for Autonomous Surface Vehicles (ASVs), their robustness in real-world conditions, particularly under external disturbances, remains insufficiently explored. In this paper, we evaluate the resilience of a DRL-based agent designed to capture floating waste under various perturbations. We train the agent using domain randomization and evaluate its performance in real-world field tests, assessing its ability to handle unexpected disturbances such as asymmetric drag and an off-center payload. We assess the agent's performance under these perturbations in both simulation and real-world experiments, quantifying performance degradation and benchmarking it against an MPC baseline. Results indicate that the DRL agent performs reliably despite significant disturbances. Along with the open-source release of our implementation, we provide insights into effective training strategies, real-world challenges, and practical considerations for deploying DRLbased ASV controllers.
As Large Language Models (LLMs) have risen in prominence over the past few years, there has been concern over the potential biases in LLMs inherited from the training data. Previous studies have examined how LLMs exhibit implicit bias, such as when response generation changes when different social contexts are introduced. We argue that this implicit bias is not only an ethical, but also a technical issue, as it reveals an inability of LLMs to accommodate extraneous information. However, unlike other measures of LLM intelligence, there are no standard methods to benchmark this specific subset of LLM bias. To bridge this gap, we developed a method for calculating an easily interpretable benchmark, DIF (Demographic Implicit Fairness), by evaluating preexisting LLM logic and math problem datasets with sociodemographic personas. We demonstrate that this method can statistically validate the presence of implicit bias in LLM behavior and find an inverse trend between question answering accuracy and implicit bias, supporting our argument.
Each year, tens of millions of essays are written and graded in college-level English courses. Students are asked to analyze literary and cultural texts through a process known as close reading, in which they gather textual details to formulate evidence-based arguments. Despite being viewed as a basis for critical thinking and widely adopted as a required element of university coursework, close reading has never been evaluated on large language models (LLMs), and multi-discipline benchmarks like MMLU do not include literature as a subject. To fill this gap, we present KRISTEVA, the first close reading benchmark for evaluating interpretive reasoning, consisting of 1331 multiple-choice questions adapted from classroom data. With KRISTEVA, we propose three progressively more difficult sets of tasks to approximate different elements of the close reading process, which we use to test how well LLMs may seem to understand and reason about literary works: 1) extracting stylistic features, 2) retrieving relevant contextual information from parametric knowledge, and 3) multi-hop reasoning between style and external contexts. Our baseline results find that, while state-of-the-art LLMs possess some college-level close reading competency (accuracy 49.7% - 69.7%), their performances still trail those of experienced human evaluators on 10 out of our 11 tasks.