Predicting Power Grid Failures with AI
![A novel contingency screening framework leverages conditional graph diffusion to efficiently identify high-risk [latex]N-k-k[/latex] outage scenarios in power networks by directly sampling from the severity distribution and utilizing a topology-aware EVGNN-trained on base-case and [latex]N-1[/latex] data-as a fast risk surrogate, thereby circumventing the computational intractability of exhaustive contingency analysis caused by combinatorial growth and eliminating the need for iterative AC power-flow simulations.](https://arxiv.org/html/2602.09461v1/x1.png)
A new approach leverages diffusion models to rapidly identify critical vulnerabilities in power systems before they escalate into widespread outages.
Science
Predicting Climate Tipping Points with Bayesian Early Warnings
![Receiver operating characteristic (ROC) estimates demonstrate the efficacy of binary classification in predicting the sign of [latex]H_2 - H_1[/latex] within time series data, with performance sustained across varying series lengths-1000, 500, and 200-and maintained whether classification relies on posterior probabilities or [latex]\tau_K[/latex] derived from sliding window estimates of the Hurst exponent with a window length of [latex]n/4[/latex].](https://arxiv.org/html/2602.09731v1/x12.png)
A new statistical framework uses time-varying patterns in climate data to identify subtle signals preceding critical transitions, offering improved foresight for long-range dependent systems.
Eyes on the Earth: Documenting Resource Extraction with AI
A new system uses artificial intelligence to analyze satellite imagery of mining sites worldwide, creating a lasting record of environmental impact.
The Chatbot Effect: How AI Companions Impact Our Psyche
New research explores the growing psychological risks and dependencies associated with using AI chatbots, drawing insights from real user experiences.
Beyond Values: How Organizations Shape AI Alignment
New research reveals that understanding the decision-making processes within organizations is crucial to building AI systems that reflect human values.
Speaking Truth to AI: Securing the Next Generation of Voice Assistants
A new framework reveals the hidden vulnerabilities in AI voice agents, highlighting the need for robust defenses against increasingly sophisticated attacks.
Beyond the Human Firewall: AI-Powered Risk Assessment
New research shows artificial intelligence can effectively scale expert judgment for evaluating complex threats, particularly in cybersecurity.
Riding the Wave: Modeling Epidemic Volatility with Stochastic Dynamics
A new theoretical framework leverages the Moyal distribution to connect microscopic transmission events with the large-scale patterns observed in pandemic outbreaks.
When Spatial Statistics Meet Deep Learning
New research reveals a surprising mathematical link between the established field of spatial statistics and the rapidly evolving world of large neural networks.
Eyes on the Sky: AI for Safer Flights
New research explores how deep learning can improve bird strike prevention by automatically identifying species and predicting flock behavior near airports.