VOR-Bench is a new benchmark for video object removal that addresses shortcomings in current evaluation methods by providing a dataset with paired edited videos and graffiti masks, a realistic motion-capable paired-video acquisition framework (rMPAF), and a perception-driven scoring model (VOR-MDSM). The dataset includes diverse data from model-generated, tool-rendered, and camera-captured sources, ensuring robust real-world assessment. Experiments show that VOR-Bench’s evaluation results correlate strongly (ρ > 0.9) with human subjective judgments, bridging the gap between traditional metrics and human preference.
By Haonan Huang, Tianrui Qiu, Xianghao Zang, Yinan Du, Zhixiang He, Chi Zhang, Hao Sun, Zhongjiang He, Tianwei Cao, Xuchong Zhang, Hongbin Sun, Kongming Liang, Zhanyu Ma
The paper investigates how to choose model pools for Multi-Agent Systems (MAS) that combine multiple model outputs to tackle complex reasoning tasks. It evaluates eight selection strategies—such as model size, accuracy, and answer diversity—across both before-generation (routing) and after-generation (majority-voting, LLM-as-a-judge) MAS architectures on scientific benchmarks. The study finds that expanding the candidate pool often harms performance, that selecting candidates within a single model family yields the best relative gains, and that indiscriminate addition of heterogeneous models can destabilize the system.
By Sara Vera Marjanovi\'c, Jiacheng Xu, Aleksandr Laptev, Grigor Nalbandyan, Erik Arakelyan, Evelina Bakhaturina
The paper introduces CTAN, a Cycle-Temporal Attention Network for audio‑visual embodied navigation. It proposes an Audio‑Visual Reconstruction Cross‑Attention module that uses bidirectional cycle‑consistency to strengthen spatial semantics across visual and acoustic modalities, and a Temporal Cross‑Modal Memory to fuse real‑time multimodal features with historical context. Experiments on Replica and Matterport3D show that CTAN outperforms prior methods in success rate, SPL, and scene navigation accuracy.
By Teng Liu, Yinfeng Yu
The paper introduces Quantum‑Harbor, a virtual laboratory that lets AI agents interact with quantum systems in a controlled setting, enabling verification of their actions and conclusions. Using this platform, the authors created QIQCBench, a benchmark of 49 expert‑authored tasks covering calibration, control, error correction, compilation, sensing, and networking. Testing 17 state‑of‑the‑art agentic systems on QIQCBench revealed wide variation in verified performance, highlighting a gap between demonstrated capability and reliable operation and positioning Quantum‑Harbor as a foundation for measuring progress toward verified autonomy in quantum engineering.
By Naixu Guo, Changhao Li, Siyu Cheng, Qicheng Tang, Binzhao Luo, Bikun Li, Yuxuan Du, Shihao Ru, Jiaqi Cai
The paper introduces a tensor-based formulation of the Viterbi algorithm for Hidden Semi-Markov Models (HSMMs), converting inner loops into tensor operations that align with SIMD and massively parallel architectures. It presents optimized implementations for single- and multi-core CPUs and, for the first time, GPUs. Experiments show speedups of up to 14× on a single core, over 200× with multi-core, and more than 570× on GPU compared to the sequential baseline, setting a new performance benchmark for large-scale HSMM decoding.
By Lorenzo Piarulli, Elia Belli, Daniele De Sensi
AsyncCouple-Flow introduces a new framework for multi‑modal spatio‑temporal forecasting that tackles three key challenges: differing sampling rates, missing modalities, and autoregressive error accumulation. It employs a Modality‑Aware Token Sparsification module to produce equal‑length sequences, an Asynchronous Cross‑Modal Coupling Graph to fuse data under arbitrary asynchrony and missingness, and a Flow‑Matching Forecasting Head that models multi‑step prediction as a conditional ODE. Experiments on weather and traffic datasets demonstrate that the method outperforms state‑of‑the‑art baselines and remains robust even when up to two modalities are missing.
By Zhixiang Wu, Yining Liu, Bo Zhao, Szu-Yu Chen, Huiran Duan, Chu Lin, Chuanguang Yang
IRENE is a deep learning model that provides probabilistic short‑range precipitation nowcasts over Italy at 1 km spatial and 5‑minute temporal resolution. It uses an encoder–forecaster architecture built on multi‑scale Convolutional Gated Recurrent Units (ConvGRUs) and is trained on national radar composites, with an importance‑sampling scheme and the almost‑fair Continuous Ranked Probability Score as its primary loss. Three training variants—standard, adversarial (IRENE‑GAN), and spectrally constrained (IRENE‑GAN‑RAPSD)—outperform benchmark methods STEPS and DGMR in probabilistic skill, though the advantage in mean absolute error is limited to the first 90 minutes.
By Alessandro Camilletti, Gabriele Franch, Elena Tomasi, Marco Cristoforetti
The paper proposes storing finite collections of knowledge as the orbit of a single compact latent generator that cycles exactly back to its starting point. By encoding each item as a phase of a fixed rotation in a learned latent space and decoding all phases with a shared network, the method guarantees exact closure through a discrete Fourier operator. Experiments on images and video clips show that this exactly periodic operator outperforms general learned or norm‑preserving operators, achieving comparable or better fidelity while eliminating visible seams and enabling efficient compression.
By Siddharth Pal, Viktoria Rojkova
ViCo is a training framework that improves the visual quality of AI-generated academic charts by using iterative self-reflection to align chart images with reference designs. It introduces a self-supervised warm-up stage that augments Monte Carlo Tree Search with consistency-based pruning, and a multi-step reinforcement learning algorithm that employs counterfactual baselines to address sparse rewards. An automatic, multifaceted evaluation framework assesses style, layout, and semantic consistency through a hierarchical heterogeneous layout graph, and experiments on three public benchmarks show ViCo’s performance approaching that of proprietary large language models.
By Jiaxin Duan, Dian Jiao Shuai Zhao, Jiabing Leng, Yiran Zhang, Feng Huang
The paper introduces EBL, an Efficient Broad Learning framework designed for distributed adaptive harmonic estimation in power grids affected by electric vehicle charging. It leverages a quantised FPGA implementation to provide high‑accuracy, half‑cycle input harmonic predictions with ultra‑low latency, outperforming existing FPGA methods by 17.4×. The online transfer learning component enables rapid adaptation across multiple charging scenarios, while bespoke quantisation and sparsity reduce resource usage to just 5.9% of the LUTs on a Zynq Ultrascale+ FPGA, compared to 82% of the state‑of‑the‑art accelerator.
By Changhong Li, Georgios Floros, Biswajit Basu, Shreejith Shanker
The paper introduces the Neverwhere Visual Parkour Benchmark Suite, a collection of over sixty hyper‑photo‑realistic 3D Gaussian Splatting reconstructions of urban indoor and outdoor scenes designed to evaluate visual locomotion controllers in closed‑loop, continuous testing setups. It aims to bridge the gap between training and real‑world evaluation by providing reproducible environments and policy checkpoints trained across multiple scenes, while highlighting the risks of relying solely on 3D Gaussian‑generated data. The authors offer code and data on their project page for easy integration into robotic evaluation pipelines.
By Ziyu Chen, Henghui Bao, Haoran Chang, Alan Yu, Ran Choi, Kai McClennen, Gio Huh, Kevin Yang, Ri-Zhao Qiu, Yajvan Ravan, John J. Leonard, Xiaolong Wang, Phillip Isola, Ge Yang, Yue Wang
The paper proposes a human-centered framework for validating the semantic soundness of machine learning models used in network traffic classification. It extends existing knowledge-generation methods by integrating data, models, explainability tools, visualizations, and expert reasoning to iteratively explore, verify, and refine model behavior and preprocessing steps. The framework is built on literature findings, benchmark analyses, XAI experience, and expert feedback, offering practical guidance for ensuring models learn trustworthy, semantically meaningful patterns rather than spurious correlations.
By Igor Cherepanov, David Sessler, Alex Ulmer, Thorsten May, J\"orn Kohlhammer
The paper introduces a collaborative optimization Boosting model for multiclass imbalanced learning that integrates density and confidence factors to create a noise‑resistant weight update mechanism and a dynamic sampling strategy. The modules are tightly coupled to coordinate weight updates, sample region partitioning, and region‑guided sampling. Experiments on 40 public imbalanced datasets show the model significantly outperforms seven state‑of‑the‑art baselines.
By Chuantao Li, Zhi Li, Jiahao Xu, Jie Li, Sheng Li
The paper introduces CoER, a framework that defends language‑model agents against adaptive indirect prompt injection (IPI) by employing attacker‑defender co‑evolution and refinement. CoER models IPI as a general‑sum Markov game, uses Co‑PPO to maintain historical opponent populations, and fine‑tunes defenders only on verified safe demonstrations. In experiments across seven domains, CoER cuts attack success from 38.5% to 0.2% while boosting task utility from 63.2% to 76.3%.
By Boyang Zhang, Qingxin Xiao, Lingwei Dang, Qingyao Wu
The paper demonstrates that the way evaluation streams are assembled in streaming intrusion‑detection benchmarks—by interleaving, pooling, or replaying network captures—acts as an uncontrolled experimental variable that can significantly alter performance metrics. In the CICIDS2017 benchmark, reordering the same set of records under a fixed split changes the held‑out samples’ overlap, prevalence, and even reverses the ranking of two deterministic scorers. Similar effects are observed in the LITNET‑2020 benchmark, where pooling disjoint captures yields a single operating point that masks large variations in per‑capture prevalences, and minor changes in batch composition can shift reported AUC‑PR values by a few thousandths.
By Michel A. Youssef
The study evaluates ten bias audit instruments across ten advanced language models on occupational gender, age, and socioeconomic status. While each tool reliably detects bias, their rankings of model performance are essentially random, indicating that different audits measure distinct constructs. The findings show that a single audit can identify bias direction within its own framework, but no audit can consistently rank models against one another.
By William Guey, Pierrick Bougault, Wei Zhang, Vitor D. de Moura, Jos\'e O. Gomes
The paper introduces ReMova, a pipeline for cleaning Belarusian data and fine‑tuning large language models (LLMs) for English‑to‑Belarusian translation. It uses a correction tool to handle the two orthographies of Belarusian, remove noise, filter out interference from other languages, and correct common misspellings found online. Ablation experiments on unfiltered data show that filtering benefits all fine‑tuned models, with LLM‑based models gaining about twice as much as a dedicated encoder‑decoder MT system, highlighting data quality as a key bottleneck for Belarusian MT.
By Mikita Pilinka, Aliaksandr Kliuje\u{u}, David Samuel, Yves Scherrer
The paper introduces the SAST-IR framework to evaluate large language models’ robustness against persuasion attacks in a memory‑less setting, revealing a flaw called "Refusal Inertia" that masks true vulnerability. Using the CP‑Agent and a custom CounterFact‑Strict dataset, the authors demonstrate that simple, diverse attack strategies achieve a 96% success rate, while complex attacks often trigger defensive compliance. The study highlights severe brittleness in current state‑of‑the‑art models when deprived of conversation history.
By Zhuoang Cai
The paper introduces GUARD, a neuro‑symbolic system that autoformalizes argumentative material by completing missing premises (guards) before formal verification. It uses large language models to generate candidate guards, Isabelle/HOL to verify them, and a contrastive test to ensure the proof depends on the original premises and does not over‑generalize. Experiments on Debatepedia and ARCT show that GUARD improves verified‑faithful scores by over 30 points and reduces leakage by about 20 points compared to prior LLM‑driven theorem proving methods.
By Xin Quan, Reto Gubelmann, Andr\'e Freitas
MechReason is a new benchmark for multi-image, multi-hop reasoning in mechanical engineering, featuring 12,000 question-answer pairs with explicit reasoning chains and 21,000 visual items across nine evidence types. It covers eight task types—explanation, prediction, design, and diagnosis—within four reasoning dimensions, and is constructed through a four-stage pipeline that extracts engineering claims, generates verification-masked questions, and validates multimodal quality. Even state‑of‑the‑art models score only 62.89% on this dataset, highlighting its difficulty.
By Tengyue Wang, Kang An, Chenxu Du, Zhongyu Yang, Yuanchi Zhu, Xinqi Yang, Hebao Zhu, Ziliang Wang, FaQiang Qian, Yunli Yang, Qibing Ren