Graph Neural Network‐Based Reinforcement Learning for Decentralized Multi‐Robot Manipulation
Robot arms lifting a large object face a trade‐off: centralized controllers explode in parameters, while decentralized ones cannot coordinate. A GNN resolves this—each arm runs its own network but acts on the full team state, achieving centralized‐level coordination with decentralized execution. Trained across team sizes, a single policy scales to four‐
Tong Chen +3 more
wiley +1 more source
STGFormer: Spatio-Temporal Graph Transformer for Traffic Flow Prediction in Sparse-Sensing Scenarios. [PDF]
Zhang J, Tan F, Bai W, Peng F, Wang W.
europepmc +1 more source
AI‐Powered Smart Displays and AR Glasses: Perspectives and Challenges
This perspective examines how AI is reshaping smart displays and AR glasses. The latest in image‐quality enhancement, sensor‐integrated display optimization, and physiological sensing in display systems, followed by AI‐assisted AR technologies including waveguide optimization, image compensation, foveated rendering, and multimodal sensing, is discussed.
Luke Benoit +8 more
wiley +1 more source
Beyond Static Representations: Learning Electroencephalogram Spatiotemporal Dynamics for Enhanced Emotion Recognition. [PDF]
Liu K, Zhu W, Xu R, Wang W, Yu Z, Wu W.
europepmc +1 more source
Bronchoscope Tip Localization With Learned Airway Constraints and an Open Source Simulator
BronchoLoc addresses feature‐poor bronchoscopic imaging by combining monocular video with a preoperative CT graph to estimate 3D camera location. A local tracker identifies candidate branches, while a sequence‐aware global module refines the trajectory. Demonstrating sub‐centimeter accuracy across synthetic and phantom datasets, this framework advances
Emanuele Nicotra +7 more
wiley +1 more source
An opinion on machine learning for predicting nanoparticles. [PDF]
Al-Zoubi W, Deeb A, Park N.
europepmc +1 more source
ZEBRA‐Prop makes language‐model‐based property prediction practical without specialized computing expertise or high‐performance computing systems. The framework predicts materials properties directly from natural‐language descriptions of crystal structures, integrating multiple descriptions through learnable weights while the language model stays ...
Ryoma Yamamoto +4 more
wiley +1 more source
Dynamic topology-aware multimodal hypergraph fusion network for load forecasting in novel power systems. [PDF]
Lv X, Ma Q, Du J.
europepmc +1 more source
DeepCPP: A Deep Learning‐Based Framework for Enhanced Prediction of Cell Penetrating Peptides
DeepCPP is a multimodal deep learning framework designed to improve the identification of cell‐penetrating peptides. By integrating sequence‐derived evolutionary information with structural graph representations through a dual‐encoder architecture and fusion strategy, the model captures complementary molecular features.
Soualihou Ngnamsie Njimbouom +2 more
wiley +1 more source
CD-TrGNN: A Complex-Domain Transformer-Graph Neural Network for ISAR Space Target Attitude Estimation. [PDF]
He Y +6 more
europepmc +1 more source

