A new large language model agent framework called LEMO Agent has been introduced to automate and accelerate the inverse design of metal-organic frameworks for gas separation.
This paper investigates how AI agents process and act upon conflicting retrieved memories during multi-step tasks, identifying vulnerabilities in their decision-making.
LOGOS is a pluggable governance and self-evolution layer designed to manage and control how multi-agent teams adapt and interact with human users over time.
This paper proposes a hierarchical, skill-based architecture for LLM agents to prevent decision-space explosion when dealing with large tool registries.
A new study suggests that uneven planning performance in large language models is driven by two distinct latent competencies rather than a single difficulty scale.
This paper proposes an epistemic state replication model to handle non-deterministic state transitions in distributed systems run by autonomous AI agents.
The authors present an auditing framework to verify whether the risk and safety claims made by trained distributional reinforcement learning agents are empirically accurate.
OmniSCS is a new framework designed to synthesize high-fidelity safety-critical scenarios for autonomous driving by allowing editable agent states and trajectories.
The authors present a systematic survey and real-world evaluation of vision-and-language navigation systems to assess their robustness in unconstrained environments.
Researchers developed a spatio-temporal transformer model to enable accurate, ECG-free dynamic coronary roadmapping during percutaneous coronary interventions.
Researchers introduced DenseAR, a generative paradigm that models autoregressive image generation as coarse-to-fine next-dense-stride predictions using a single-scale tokenizer.
Researchers proposed an exponential-linear weight reparameterization method to improve neural network optimization by addressing the multiplicative nature of weight updates.
Researchers propose modeling latent token sequences in hidden-space reasoning methods as dynamical systems to improve the interpretability of latent Chain-of-Thought models.
The authors introduce a framework to measure and ensure the rank stability and precision of domain citations in generative search engine visibility metrics.
Zeekr's 7GT vehicle achieved a five-star safety rating under 2026 Euro NCAP standards, utilizing Bosch's proprietary intelligent driver assistance system.
Omni-Decision is a training-free agent system designed to track and ground scattered multimodal evidence for complex question-answering tasks.
A new lightweight, audio-only baseline detects Alzheimer's disease from spontaneous speech recordings using handcrafted acoustic biomarkers instead of relying on transcripts or heavy deep learning models.
China Merchants Securities advises investors to focus on domestic AI computing opportunities, highlighting Huawei's upcoming showcase of its Atlas 950 SuperPoD.
Shanghai Electric projects a 12% to 22% year-over-year increase in net profit for the first half of 2026.
Thirty business leaders from various Chinese and international companies have shared their perspectives on the future of AI development in China ahead of the 2026 World AI Conference.
Alibaba Cloud is updating the pricing and minimum purchase requirements for its AgentOne intelligent calling service packages.
Data from the China Association of Automobile Manufacturers shows that Tesla and SAIC were the only top-ten sedan manufacturers to see sales growth in the first half of 2026.
This paper conceptualizes the iterative interaction loop between humans and AI agents as a high-frequency training environment that can induce cognitive and neuroplastic changes in users.
Researchers have proposed dynamic resource allocation methods to enhance Ensemble Determinization Monte Carlo Tree Search in highly uncertain and adversarial environments.