FlyMark is a training‑free invisible watermarking method for 3D Gaussian Splatting (3DGS) scenes that embeds a keyed message directly into the scene’s parameter array. It uses directions derived from a published fruit‑fly connectome to encode bits via a dithered quantization‑index‑modulation scheme, achieving high clean bit accuracy and visual fidelity while preserving all geometry and appearance parameters exactly. Extraction requires only simple cone queries, rounding, and majority voting, and the watermark remains robust under a model‑domain threat model on both synthetic and real scenes.
By Ziyuan Luo, Haoliang Li, Renjie Wan
The paper proposes that super‑resolution should be performed in a latent space that is close to the manifold of high‑resolution images, rather than in reconstruction‑oriented or pixel spaces. By using the latent space of 23 fused layers of a frozen DINOv3‑L model, degraded images are embedded near this manifold, enabling a 415M decoder (RAESR) to map them back and decode to pixels in a single pass. RAESR achieves the best fidelity–perception trade‑off on several benchmarks while running at 37 ms per 512×512 image on a single H20 GPU, outperforming models that use a VAE latent substrate.
By Wanzhou Lei, Cuifeng Shen, Yanjin He, Maohua Li, Hua Yuan, Per-Olof Persson, Tao Lan, Kan Liu, Hanlin Tang
The paper introduces PARCEL, a benchmark designed to evaluate whether legal claims are supported by cited authorities. Using 3,396 claims from New York State Court of Appeals decisions, the authors label each claim as Supported, Refuted, or Not Found and cast the task as a three‑way natural language inference problem. State‑of‑the‑art LLMs achieve up to 0.97 accuracy in zero‑shot settings, yet they still frequently misclassify unsupported claims as supported, especially when fabricated but plausible citations are present.
By M. Mikail Demir, M. Abdullah Canbaz
The paper investigates the large‑sample variational problem for t‑Distributed Stochastic Neighbor Embedding (t‑SNE) using a broad class of input and output kernels. It shows that an entropy equation uniquely determines the scale parameter of the input kernel at interior points of positive density, and provides sufficient conditions for the existence and uniform boundedness of solutions across the entire support. Under these conditions, together with a decay assumption on the output kernel, the discrete optimal values converge to a continuum minimum, and empirical measures of approximate minimizers are tight after translation, yielding compactly supported minimizers that satisfy the equilibrium equation. The study includes Gaussian and, in two dimensions, Cauchy output kernels, and presents numerical comparisons of two‑dimensional representations obtained with different kernels.
By Yi Gu, Antonio Auffinger
The study evaluates a pipeline for detecting dysarthria during awake craniotomy using the DATABRASE corpus. It combines speaker diarization, a multi‑view representation of handcrafted acoustic features and wav2vec 2.0 embeddings, speaker‑conditional normalization, feature selection, and a cascaded classifier. Results show that the speech representation has a larger impact on cross‑speaker performance than classifier choice, with self‑supervised embeddings improving AUC by 18.2%–26.1% and additional gains from diarization‑conditioned extraction and the classifier cascade.
By Kanthila Chinmayi (IRDL, LaTIM), Abdallah Nassib (LARIS), Misy Harrison (LaTIM), Panheleux Celine (LaTIM, CHU - BREST), Saliou Vanessa (CHU - BREST), Seizeur Romuald (LaTIM), Dardenne Guillaume (LaTIM)
The paper presents a reference‑free framework for detecting hallucinations in Sinhala‑to‑English neural machine translation. It introduces a 45,000‑sample synthetic dataset created via five linguistically motivated corruption strategies and a semantic rescue mechanism that uses character‑level similarity. Using a fine‑tuned mDeBERTa‑v3 model, the authors achieve a token‑level F1 of 0.841 and demonstrate that the detector relies on the Sinhala source, not on surface artefacts, with an AUROC of 0.970 on a source‑disjoint test set. The study also benchmarks eight NMT systems, showing that hallucination rates vary by an order of magnitude across architectures.
By Navam Obeysekara, Nevidu Jayatilleke
SciTBERT is a family of BERT-derived language models trained on chronologically consistent corpora of scientific papers, patents, and educational web text, with training cutoffs spanning 2013 to 2025. The models are further post‑trained using paper and patent citations to create the SciTBERT‑CI family, which better captures temporal dynamics. Experiments show that these chronologically consistent encoders outperform earlier domain‑specific models, especially when training data is limited by early year restrictions, and they perform well on the new PatRepEval benchmark of patent‑related tasks at the science‑technology interface.
By Thomas Gebhart, Russell J. Funk
The paper proposes augmenting the global [CLS] token and register tokens of vision transformers with spatial region tokens to better capture fine‑grained image details. By matching each semantic token with a corresponding image patch, the method automatically generates localized ROI tokens without external annotations. These tokens are then used in a multi‑vector retrieval framework inspired by ColBERT, improving fine‑grained image retrieval while keeping storage costs manageable.
By Ian de Holanda Cavalcanti Bezerra, Vivek Trivedy, Lucas Pascotti Valem, Longin Jan Latecki
The paper introduces EDMD-kDL, a kernel-based method that learns finite-dimensional Koopman embeddings directly from data using bilevel optimization and collocation techniques. It combines kernel dictionary learning with extended dynamic mode decomposition to produce Koopman approximations without pre-specifying a dictionary. Experiments on sea-surface-temperature forecasting and video data show that EDMD‑kDL matches or surpasses ANN-based approaches while remaining scalable to large datasets.
By Joel-Pascal Ntwali N'konzi, Feliks N\"{u}ske, Stefan Klus
The paper introduces GloFND, an optimization-based method that learns a threshold for each anchor image during training to automatically detect false negative pairs in self-supervised contrastive learning. Unlike prior approaches that only consider false negatives within a mini-batch, GloFND globally identifies them across the entire dataset while keeping per-iteration computation independent of dataset size. Experiments on image and image-text data show the method’s effectiveness, and the implementation is publicly available.
By Vicente Balmaseda, Bokun Wang, Ching-Long Lin, Tianbao Yang
The paper introduces PT‑RAG, a retrieval‑augmented generation framework that enhances deep learning models for predicting cellular transcriptional responses to gene perturbations. PT‑RAG adds a two‑stage retrieval module—semantic retrieval followed by a Gumbel‑Softmax selector—to supply contextually relevant perturbation information, improving predictive accuracy across cell types and perturbations. Experiments on STATE‑style and scGPT backbones show consistent gains, such as a 5.9% reduction in 2‑Wasserstein distance for cross‑cell‑type predictions.
By Andrea Giuseppe Di Francesco, Andrea Rubbi, Rishabh Jain, Pietro Li\`o
The paper introduces Efficient Multi-Granularity Knowledge Transfer (EMGKT) for radiology report generation, aiming to improve medical knowledge extraction from X‑ray images. EMGKT encodes global knowledge via a medical vision‑language model and employs Fine‑Grained Knowledge Distillation (FGKD) to learn fine‑level embeddings without extra priors during inference. Additionally, a mixture of disease‑diagnosis expert classifiers is used to further enhance knowledge extraction, and the method is shown to be effective and transferable across datasets and baselines.
By Xubin Zhong, Zheyu Zhang, Wenjian Qin, Ning Wen
ShapeY is a benchmarking framework that evaluates shape-based recognition in object recognition systems using 68,200 grayscale images of 200 3D objects rendered from multiple viewpoints, with optional appearance changes. It employs a nearest‑neighbor matching task to probe how well an embedding space clusters views by 3D shape similarity across varying viewpoints and non‑shape alterations, providing quantitative and qualitative performance readouts such as error rates, viewpoint tuning curves, and matching score histograms. Testing 321 pre‑trained networks shows that even state‑of‑the‑art models struggle to generalize robustly across 3D viewpoint and appearance changes, highlighting gaps in current shape recognition capabilities.
By Jong Woo Nam, Amanda S. Rios, Bartlett W. Mel
The paper introduces CoCo, a conjoint framework for deep graph clustering that emphasizes compactness and consistency in node representations. CoCo uses graph convolutional filters to learn robust embeddings from both local and global views, then encodes them into low‑rank compact embeddings to reduce redundancy and noise. A consistency learning strategy further enriches node semantics, enabling knowledge transfer between perspectives and yielding superior performance over state‑of‑the‑art methods on multiple datasets.
By Wei Ju, Siyu Yi, Kangjie Zheng, Yifan Wang, Ziyue Qiao, Li Shen, Yongdao Zhou, Xiaochun Cao, Jiancheng Lv
arXiv:2610. 11584v1 Announce Type: cross Abstract: To test conditional independence of $X$ and $Y$ given a text or an image $Z$, one conditions on an embedding $\psi(Z)$ in place of $Z$.
By Nikolaj Thams, Anton Rask Lundborg
4D-GSW is a kinematic-aware watermarking framework for 4D Gaussian Splatting that embeds robust copyright information while preserving high spatio-temporal consistency. It introduces a Spatio-Temporal Curvature metric to identify dynamic instants and gates watermark gradient injection, and uses a joint HMM-MRF energy minimization model to synchronize watermark phases across temporal trajectories and spatial neighborhoods. An anisotropic gradient routing mechanism ensures watermark embedding remains decoupled from photometric reconstruction fidelity, achieving superior robustness and rendering quality in experiments.
By Sifan Zhou, Hang Zhang, Yuhang Wang, Ming Li
WorldFact-Bench is a new benchmark that tests image‑world consistency by presenting 1,274 real‑fake image pairs across four verification regimes and ten semantic domains, each pair containing a factual conflict while keeping the image visually plausible. The evaluation requires both images in a pair to be classified correctly, and the benchmark focuses on single‑image assessment without predefined claims. The authors also introduce PERSIST‑Agent, an iterative verification system that maintains a persistent state linking candidate facts, visual observations, evidence, and verification statuses, and shows improved pair accuracy over baseline methods on 8B backbones.
By Zhuohong Chen, Zhengxian Wu, Yunyao Yu, Hangrui Xu, Zijian Yu, Hao Tan, Zhifang Liu, Peng Jiao, Jun Lan, Haoqian Wang
The paper introduces BATok, a budget‑adaptive tokenizer that converts raw I/Q electromagnetic signals into compact tokens suitable for multimodal large language models. BATok adjusts token capacity based on input length and signal content, using multi‑resolution features and a local energy prior to resample representations. The authors also present EMSpec‑Instruct, a dataset linking raw signals, waterfall images, and language instructions for tasks such as modulation recognition, structured detection, and signal grounding, and demonstrate that BATok achieves competitive performance across these tasks.
By Lei Zhai, Zhihao Chang, Shuyuan Yang, Zhixi Feng
SPERA is a new EEG foundation model that predicts in latent space using a joint-embedding predictive architecture. It incorporates a Legendre-polynomial spatial prior into attention to handle varying scalp electrode geometries, and adds factorized temporal and spatial attention with periodic full-attention blocks, plus a relational spectral regularizer to align latent similarity with spectral views. Trained on about 80,000 hours of EEG from 29,048 subjects across 106 datasets, SPERA achieves the highest average balanced accuracy on nine downstream tasks in clinical, cognitive, and BCI domains, while demonstrating strong parameter efficiency and robustness to different recording conditions.
By Minsu Kim, Ye-Sung Kim, Hyeseong Jeon, Wooseok Hyung, Joshua Lee, Chang-Hwan Im
The study reviews 25 years of automated verbal deception detection research, analyzing 3,653 models across 97 datasets. Pooled accuracy is 74.4% (95% CI: 71.2%-77.4%) with high heterogeneity, driven more by methodological quality than model complexity. Only a small fraction of studies use verifiable ground truth or independent evaluation, and the accuracy ceiling of 70‑75% appears unlikely to improve under current practices.
By Riccardo Loconte, Jonas Festor, Zane Fatjanova, Mariam Bolkvadze, Bennett Kleinberg