Science

A new approach combines the broad analytical power of artificial intelligence with human judgment to proactively identify and mitigate systemic risks posed by rapidly evolving AI technologies.

A new approach uses real-time data and statistical modeling to quantify the probability of entering a period of significant market stress.

Researchers are pushing the boundaries of efficient AI model training by strategically reducing computational load at the neuron level.

A new approach unifies detection of both security vulnerabilities and unreliable outputs in advanced machine learning models.

Researchers have developed an automated pipeline to forecast the magnetic field structure of coronal mass ejections, offering a crucial step towards improved space weather prediction.

Researchers are leveraging generative machine learning to dramatically increase the scope of climate model ensembles, offering a path toward more robust and reliable predictions.

A new approach uses the power of large language models to detect subtle anomalies in system logs, offering a more proactive and adaptable defense against modern cyberattacks.

Uber engineers detail how AI-powered fault injection is proactively identifying and resolving vulnerabilities in their massive mobile infrastructure.

Researchers have identified a key mechanism driving training failures in large language models and developed an optimizer to address it.

A new analysis reveals that Low Earth Orbit satellite networks harbor unexpected vulnerabilities that standard network metrics fail to detect.