Artificial intelligence, machine learning, computer science and mathematics.
Researchers present a modular algorithm for preparing general mixed quantum states on quantum computers. Tests covered two-qubit examples and random states involving one, two and three qubits, but the abstract does not report performance measures or scaling limits.
AI systems can generate proof sketches and formal steps, but those outputs often vary in how much evidence they actually carry. This paper proposes Proof Engine Infrastructure, an architecture that links each claim to a checking procedure and composes only earned results into larger proof goals.
A study of female university students living in dormitories in Baghdad links social media use to worse mental health, including more anxiety, depression, and social isolation. The researchers report that longer daily use and entertainment or social comparison content were associated with these mental health indicators.
Researchers report DeepFakeBuster, an AI system for detecting deepfake images that combines multiple detection signals and adjusts their influence using confidence estimates. On a test set of 192,000 authentic and manipulated images, the approach reached 97.8% accuracy under the evaluated conditions.
People who constantly check their phones during face-to-face conversations—often called “phubbing”—may be harming relationships and mental health. A systematic review of 22 studies links phubbing to lower relationship quality and outcomes such as loneliness and depression, with people’s perceptions of being snubbed playing a key role.
A study of 13,800 firm-years examined how companies’ required cybersecurity disclosures vary by topic and where they appear. It found that risk-factor write-ups were more consistently linked to next-year profitability and valuation than similar text in management’s discussion section.
Researchers built a four-core optical computing chip that processed tasks across 124 channels in parallel, reaching a speed of 65 TOPS. Using the chip in an optoelectronic convolutional neural network, they reported 95.08% accuracy on MNIST, a 9.20% improvement over a single-core version.
