Weathering the Storm: Predicting Power Outages with AI
A new machine learning framework leverages weather patterns and socio-economic data to forecast power disruptions caused by extreme events.
Science
Fortifying AI: A Proactive Defense Against Emerging Threats
As artificial intelligence systems become increasingly integrated into critical infrastructure, securing them against adversarial attacks and data manipulation is paramount.
Leveling the Playing Field: Machine Learning for Rare Conditions in the ICU
Predictive modeling in emergency and critical care often struggles with limited data for the most serious-but least frequent-conditions.
Beyond Pairwise Connections: A New Network Robustness Predictor
Researchers have developed a novel approach to assessing the resilience of complex networks by modeling higher-order interactions, offering a more accurate prediction of cascading failures.
When AI Assistance Backfires: Why Teaching Humans to Correct AI Isn’t Always Effective
New research reveals that simply pairing humans with large language models doesn’t guarantee improved accuracy, highlighting critical flaws in current AI-integration training pipelines.
Fighting Fraud with Synthetic Data: A New Approach to Collaborative Detection
A novel method for generating privacy-preserving synthetic financial datasets enables more effective and explainable fraud detection through collaborative data sharing.
Staying Ahead of the Flood: Smart Leak Prediction for AI Data Centers
![Data streams from networked sensors converge upon a dual-model machine learning system-an [latex]LSTM[/latex] for forecasting and a Random Forest for anomaly detection-with results surfaced through a real-time Streamlit dashboard, establishing a closed-loop system for monitoring and preemptive alerts.](https://arxiv.org/html/2512.21801v1/leak_detection_architecture.drawio.png)
A new IoT framework leverages machine learning to forecast and detect leaks in liquid cooling systems, minimizing downtime and maximizing energy efficiency.
Decoding the Digital Underworld: Predicting Cyberattacks with Telegram Insights
A new framework analyzes public Telegram channels to identify emerging cyber threats before they materialize, offering a proactive defense against malicious actors.
Governing the AI Pipeline: A Framework for Secure NLP Models
As natural language processing becomes increasingly integrated into critical systems, organizations need robust protocols to ensure these models are secure, compliant, and reliable.
Decoding Market Shifts with Machine Learning
![The correlation structure reveals a shared underlying pattern between synthetically generated data-built upon a [latex]Gaussian-Bernoulli[/latex] model-and data originating from real-world observations, suggesting the model effectively captures essential relationships present in the observed phenomena.](https://arxiv.org/html/2512.21823v1/gaussian_corr.png)
A new approach leverages conditional Restricted Boltzmann Machines to identify structural changes in financial time series beyond traditional volatility analysis.