Researchers introduced SANDBOXESCAPEBENCH, a benchmark to evaluate the security of large language models in sandbox environments, revealing that LLMs can exploit vulnerabilities to escape isolation. T…
Researchers developed PRML2, a hybrid framework combining Kalman filtering with machine learning, to improve vehicle localization accuracy in autonomous driving, especially when GPS signals are weak. …
Current large language models like GPT-5 and Claude Opus-4.1 achieve only a 40% success rate on the PROBE benchmark for proactive AI agents, revealing significant limitations in autonomous problem-sol…
Supervised Reward Inference (SRI) offers a new approach to AI goal understanding by mapping observed behaviors to rewards, outperforming traditional methods in flexibility and accuracy. SRI achieves n…
Quantum kernels could revolutionize machine learning by encoding data into quantum states, but they face design challenges like exponential concentration. A new approach called Local-Global quantum ke…
New research reveals that the L0 hyperparameter in sparse autoencoders (SAEs) is critical for feature disentanglement in large language models, with improper tuning leading to mixed or degenerate feat…
Researchers developed a robotic throwing method using potential field state representation that achieves up to 90% success rates in cluttered environments with unseen objects. The approach, trained vi…
Researchers developed a controllability-observability framework to reduce hidden-state redundancy in deep neural networks, achieving up to 72.95% state compression on MNIST with minimal accuracy loss …
Researchers developed a deep reinforcement learning framework based on Proximal Policy Optimization to optimize battery charging for autonomous mobile robots in warehouses. The model dynamically selec…
SHARC, a new framework for SHapley Additive exPlanations for Regulatory Capital, makes Gaussian Process Regression models auditable by decomposing Stressed Value-at-Risk into baseline, mean-driven, an…
A new policy called UCB-DPP optimizes the balance between automated chatbots and human agents in hybrid service systems, achieving regret of approximately K√T and ensuring mean-rate stability in human…
AI models struggle in real-world negotiations because they focus on language rather than strategic economic decision-making. A new training method called Reinforcement Learning from Verifiable Rewards…
New research introduces a framework called Statistically Meaningful Geometry (SMG) to evaluate whether large language models (LLMs) are genuinely intelligent or merely pattern-matching. The framework …
AI uncertainty in predictions raises legal and ethical issues, with selective abstention and friction as proposed interventions. Studies show uncertainty-based abstention may worsen disparities for un…
Researchers introduced a Gibbs algorithm that simplifies posterior sampling in Bayesian inversion, offering reliable convergence guarantees and computational efficiency. The method, verified through n…
Researchers introduced CcGAN-AVAR, an extension of the Continuous conditional Generative Adversarial Network, to address label imbalance and improve efficiency in continuous conditional generative mod…
Researchers are exploring retrieval-augmented generation (RAG) and constrained decoding (CD) to improve large language models' ability to generate accurate web API integration code. While RAG reduces …
Onnes, a digital twin for dilution refrigerators, uses AI and LLM agents to improve fault diagnosis in quantum computing hardware, achieving 100% recall and near-perfect classification accuracy with m…
New smooth Monte-Carlo estimators can evaluate the Euler characteristic in continuous neural fields 250 times faster than traditional persistent-homology methods, but they require strict design rules …
A study of 692 publications reveals that AI fairness research is heavily concentrated in the United States, with low- and middle-income countries largely absent. The skewed citation influence suggests…