The paper implements two quantum graph neural network architectures—Simplified Graph Convolution (SGC) and Linear Graph Convolution (LGC)—and evaluates them on benchmark graph datasets for semi‑supervised learning using quantum simulation. It compares their predictive performance and optimization behavior to classical baselines, finding that the quantum models achieve competitive results with fewer parameters. Additionally, the study provides a cost‑gradient analysis to identify trainable tasks and a classical simulability investigation to determine regimes where the circuits remain robust during training.
By Paul San Sebastian Sein, Theodor Iosif, Tilen G. Limb\"ack-Stokin, Kin Ian Lo, Yidong Liao
TinyCNN is a lightweight convolutional neural network with only 193,190 trainable parameters designed for on‑device plant disease classification. It achieves 98.88% test accuracy on the 38‑class PlantVillage benchmark, outperforming larger models while consuming far less energy, memory, and cost. The study also explores knowledge distillation to further compress the model and evaluates cross‑dataset robustness, finding a significant performance drop when moving from PlantVillage to PlantDoc due to background‑driven shortcut learning.
By Ngoc-Bao Ho-Lam, Thai-Anh Nguyen
The paper studies Online Kernel Supervised Principal Component Analysis (OKSPCA), which uses random features and an Adam-style orthonormal basis update to optimize a supervised spectral objective. It shows that accurate optimization of this objective does not guarantee accurate population subspace recovery or improved predictive performance, and it provides theoretical results on consistency, concentration, and perturbation of the estimator. Empirical experiments on six benchmarks reveal that replacing the tracker with the exact empirical target does not significantly change regression deficits, while classification-rank models capture most of the terminal objective energy but can exhibit substantial geometric deviation; sample-size studies further separate empirical accuracy from population recovery. The diagnostics also compare computational trade-offs, indicating that exact on-request computation can be faster in classification settings, whereas Adam saves time relative to full thin‑SVD in some dense regression requests, despite persistent geometric error.
By Zhenlin Yao, Wei Xiong
MTF‑Net is a Multi‑Modal Temporal Feature Fusion Network that jointly models kinematic, appearance, and contextual cues for pedestrian intention prediction. It fuses four modalities—bounding‑box dynamics, human pose keypoints, local context, and scene‑level semantics—within a recurrent framework enhanced by gated linear units (GLUs) and an attention‑guided fusion head. Evaluations on the PIE and JAAD benchmarks show that MTF‑Net outperforms recent transformer‑ and graph‑based models, achieving up to 0.95 AUC on PIE and 0.94 AUC on JAAD while maintaining real‑time performance.
By Md Mahfuzur Rahman, Pengzhan Zhou, A. F. M. Abdun Noor, Md Imam Ahasan, Md Mustafizur Rahman, Fang Qu
The paper introduces TSMSA-Net, a two‑stage multi‑scale attention‑based network designed for weakly supervised enhancement of cataract fundus images. It first synthesizes paired real‑like cataract images via domain transformation, then applies a multi‑scale attention stage to extract detailed features while minimizing artifacts. Experiments on Kaggle and ODIR‑5K datasets show that TSMSA‑Net surpasses current state‑of‑the‑art methods even without paired data and improves vessel segmentation and classification performance.
By Xiaoyong Fang, Yue Wang, Xiangyu Li, Wanshu Fan, Dongsheng Zhou
The paper introduces Voyager, a lightweight router that learns to select a sparse, final‑layer‑anchored expert from CLIP’s layer depth for out‑of‑distribution (OOD) detection. Using a weak OOD supervision context, Voyager outperforms state‑of‑the‑art prompt‑learning methods on three benchmarks, achieving an average FPR@95 of 18.86 on ImageNet‑1K. The router can be trained in about two minutes with less than 1 GB of memory, making it roughly 20× more efficient than existing approaches.
By Ignacio M. De la Jara, Cristian Rodriguez-Opazo, Damith Ranasinghe
The paper evaluates three compact computer‑vision models—each with about 0.26 million trainable parameters—for iris presentation attack detection (PAD) under conditions not seen during training, such as new presentation attack instruments (PAIs) and environmental degradation. The models are trained from scratch without external pretraining or data augmentation and tested on the Notre Dame subset of LivDet‑Iris 2017 across known‑attack, unknown‑attack, corrupted, and pooled partitions. Results show significant performance drops when moving from known to unknown attacks, with the best model, ZACH‑ViT, achieving the lowest unknown‑attack APCER (≈47.7 %) and D‑EER (≈38.9 %), yet overall errors remain high enough to preclude deployment readiness.
By Athanasios Angelakis, Marta Gomez-Barrero
FunArt is a framework that builds articulation‑aware functional 3D scene graphs from a single static RGB‑D observation. It reconstructs object instances, converts their geometry into the O‑Voxel representation of TRELLIS.2, and uses a frozen sparse‑compression VAE as a structural prior. A lightweight query‑based decoder jointly segments movable parts and functional interactive elements while estimating motion type, axis, origin, and range, achieving state‑of‑the‑art results on the Articulate3D dataset.
By Dennis Rotondi, Abdelrhman Werby, Kai O. Arras
G^2RA‑Net is a medical image segmentation framework that models cross‑slice relations using a graph-based approach and refines segmentation with a cross‑slice attention gate. It represents each image slice as a graph node, propagates semantic context through graph message passing, and then selects relevant neighboring context to emphasize target anatomical regions. Experiments on brain MRI and abdominal CT datasets show that G^2RA‑Net improves segmentation accuracy and boundary quality compared to representative methods, with ablation studies confirming the effectiveness of its design.
By Shengye Wang, Zonglin Wu, Liang Fan, Yule Xue, Haozhe Zhao
SplashSplat introduces a new benchmark of 20 real-world splashing liquid scenes captured with seven synchronized 4K cameras at 60 fps, providing per-view liquid and container masks and fixed evaluation splits. The method reconstructs per‑frame liquid signed distance fields (SDFs) from these masks, fuses them into a coarse velocity field, and uses Lagrangian carriers to generate differentiable local Gaussian representations for rendering. SplashSplat outperforms existing dynamic Gaussian splatting techniques on both real and synthetic data, offering more physically plausible motion, lower training cost, and enabling temporal interpolation and style transfer without re‑optimization.
By Peiyu Liu, Dingxi Zhang, Federico Tombari, Marc Pollefeys, Christina Tsalicoglou, Daniel Barath
UniReg is a conditional unified model for medical image registration that adapts deformation field estimation based on anatomical priors, registration type constraints, and instance-specific features. It combines the precision of task‑specific learning with the generalization of traditional optimization, enabling effective alignment across diverse CT and MR scenarios within a single framework. Experiments show UniReg outperforms state‑of‑the‑art learning‑based methods in accuracy while providing strong cross‑scenario generalization and reducing training cost and model redundancy.
By Zi Li, Jianpeng Zhang, Tai Ma, Tony C. W. Mok, Yan-Jie Zhou, Zeli Chen, Xianghua Ye, Le Lu, Cheng Chen, Dakai Jin
SceneTeract is a verification interface that separates semantic action understanding from physical feasibility in indoor 3D scenes. It decomposes activities into atomic actions and performs explicit geometric checks to determine executability, providing diagnostic traces for failures. The system reveals widespread functional and accessibility issues in synthetic scenes, shows that existing VLMs over‑predict action feasibility, and improves VLM performance through post‑training with verifier feedback, with benefits that generalize to real‑world scenes.
By L\'eopold Maillard, Francis Engelmann, Tom Durand, Boxiao Pan, Yang You, Leonidas Guibas, Maks Ovsjanikov
The paper introduces Structured Four-Stage Legal Translation (S4L→Prolog), a reasoning-guided framework that converts raw traffic rules into Prolog logic by performing semantic role extraction, scene completion, logical mapping, and rule generation in a single prompt. Compared to baseline approaches (NL→Prolog and LE→Prolog), S4L achieves higher accuracy—formalizing 75 % of twenty real-world traffic rules versus 60 % and 55 % for the baselines. Qualitative analysis shows S4L better captures implicit causal relations, deontic modality, and exception structures.
By May Myo Zin, Wachara Fungwacharakorn, Ken Satoh, Katsumi Nitta
RAFT is a retrieval‑augmented framework designed for troubleshooting agents that treats support cases as stateful, multi‑stage documents. It represents each closed case as a directed chain of timeline entries and retrieves at the entry level, matching intermediate states of an active case and returning the parent‑case trajectory anchored at the matched state. The authors evaluate RAFT on a synthetic benchmark and real Apache Jira issues, showing statistically significant improvements over vanilla RAG and GraphRAG baselines, and release the benchmark, implementation, and evaluation set.
By Mingxuan Zhang, Xiaowen Wang, Anupma Sharan, Zhengyi Chen, Chenyu Diana Zhang, Shanshan Yang, Chittibabu Pacharu
LinePilot Digitizer is a tool for extracting numerical data from line plots, offering three calibration modes—standard, enhanced, and OCR—that combine continuous color-based curve recovery with flexible calibration. The authors introduce DigitizerBench, a benchmark that systematically evaluates digitizer performance across signal, rendering, and plot-structure factors, using a failure-penalized capped normalized root-mean-square error metric. Results show that LinePilot (OCR) and LinePilot (enhanced) achieve the lowest error rates and highest trusted usability on their respective benchmark subsets, demonstrating the effectiveness of the calibration modes and the benchmark’s factor analysis capabilities.
By Fengbo Ma, Rayan Akhtar, Aakash H. Joshi, Xiaoting Li, Haijian Sun, Zhen Xiang, Xianyan Chen, Yiping Zhao
The paper presents a method for dense univariate time‑series segmentation by treating each time point as a node in a graph and applying a Graph Attention Network (GAT). Using visibility‑based graph constructions, particularly WDPVG, the approach achieves competitive performance on a 59‑dataset benchmark, matching or surpassing traditional point‑wise classifiers while requiring far fewer parameters and less memory. The method’s advantage stems from long‑range edges that provide contextual information beyond local windows.
By Ivana Kesi\'c, Carolina Fortuna, Mihael Mohor\v{c}i\v{c}, Bla\v{z} Bertalani\v{c}
The paper analyzes the environmental footprint of machine learning model training, focusing on large language models and their hardware. It finds that energy use and environmental impacts have risen exponentially over the past decade, even when employing carbon‑efficient electricity and more efficient hardware. The study argues that optimization strategies alone cannot curb these impacts due to a rebound effect, and stresses the need to evaluate hardware life‑cycle impacts and integrate environmental metrics into NLP research practices.
By Cl\'ement Morand (STL), Anne-Laure Ligozat (ENSIIE, LISN, STL), Aur\'elie N\'ev\'eol (STL, LISN)
The paper introduces an adaptive Mixture-of-Experts (MoE) framework for time series forecasting that incorporates expert-specific losses to give each expert a direct learning signal independent of gating weights. The overall objective combines base forecasting loss with these expert losses, encouraging experts to specialize on different temporal segments. A partial online learning strategy is added for efficient incremental updates, and experiments on economic, tourism, and energy datasets show the method outperforms state‑of‑the‑art neural models and foundation models, with ablation studies confirming the benefit of expert loss integration.
By Btissame El Mahtout, Florian Ziel
The paper introduces TrioRAG, a graph-free multimodal retrieval-augmented generation framework that combines evidence from the question, an anchor image, and a VLM-enhanced query via late fusion. It also presents AutoQA, a benchmark featuring noisy web-sourced images that require reasoning across manuals. TrioRAG outperforms graph-based systems on three benchmarks while cutting costs and speeding up inference by 1.6–2.3×.
By Tithi Rakshit, Hongkuan Zhou, Lavdim Halilaj, Yuqicheng Zhu
The paper introduces pre‑trained models for extracting variant‑phenotype relations from biomedical text, focusing on the SNPPhenA corpus. Fine‑tuning small BERT‑based models, especially DeBERTa, achieves performance close to the current state‑of‑the‑art. Moreover, careful fine‑tuning of Google’s Gemini Pro 1.0 surpasses existing benchmarks on both sentence‑level and abstract‑level relation extraction tasks.
By Claudiu Creanga, Liviu P. Dinu, Daniela Gifu