This survey paper reviews the current state of in-context reinforcement learning, focusing on how pretrained decision models handle non-stationary environments without parameter updates.
This research provides a theoretical framework to explain the mechanics behind data mixing scaling laws in multi-domain model training.
A new hybrid transformer architecture is presented for automating cervical cancer radiotherapy contouring, demonstrating clinical deployment capabilities.
This study critiques current evaluation protocols for AI pentesting agents, arguing that existing benchmarks fail to accurately predict performance in real-world security environments.
The SLEUTH framework improves multi-hop reasoning in language agents by explicitly managing epistemic working memory to prevent context dilution during long reasoning chains.
A new speculative decoding strategy for Mixture-of-Experts models optimizes inference efficiency by accounting for the computational costs of specific expert activations.
The GaitSpan framework enables humanoid robots to transition between walking and running gaits without requiring separate training for each movement.
The authors present a deep reinforcement learning approach to optimize the scheduling of cryptographic audits for off-chain data storage.
A field study evaluates visitor acceptance of an autonomous multilingual android robot deployed in a public museum.
DMFNet is a proposed dual-backbone network architecture aimed at improving the classification of complex urban scenes in remote sensing imagery.
A new position paper argues that explanation stability in machine learning models is tied to the specific model and interpretation method pair rather than the model alone.
This paper explores the application of AI to analyze urban mobility data for the purpose of improving transportation management in smart cities.
A novel semi-supervised learning method called OFD-Net is introduced to improve medical image segmentation by utilizing orthogonal feature disentanglement without teacher models.
This article provides a summary of key insights and discussions from industry leaders and researchers at the World Artificial Intelligence Conference.
Baidu's general-purpose AI agent, Baidu Dazi, has been recognized at the World Artificial Intelligence Conference 2026 for its productivity and task-automation capabilities.
This article explores the technical complexities and challenges involved in implementing model routing systems for AI applications.
Ferrovial is investing one billion euros to develop a new data center campus in Madrid with an initial 60MW capacity.
China Mobile has invested in Youji Technology to advance the development of AI-integrated industrial infrastructure.
MirrorSense showcased new hardware and software tools at WAIC 2026 aimed at integrating spatial perception and robotics for industrial applications.
Researchers have developed Holi-Spatial, an AI method for 3D automatic annotation that significantly improves detection accuracy without the need for manual labeling or expensive hardware.
Researchers propose a theoretical framework to address performance saturation in closed-loop AI systems by identifying how external information can overcome internal feedback limitations.
The CoreForge project demonstrates an iterative workflow where large language models are used to construct a MaxSAT solver directly from academic literature rather than existing codebases.
Researchers introduced a benchmark designed to evaluate the accuracy and traceability of LLM agents performing complex structural engineering workflows.
This paper models the value of connectivity within networks of AI agents to determine optimal collaboration protocols.