Beyond Resolution: Neural Networks That Understand Scale
New research demonstrates how incorporating scale invariance into neural network design enables robust extrapolation to unseen data scales, unlocking better modeling of self-similar phenomena.
Science
Forecasting Cyberattack Surges: A New Approach to IDS Alert Analysis
Researchers have developed a method to predict escalating periods of high-intensity network intrusion attempts by analyzing trends in security alert streams.
Beyond Bifurcations: Predicting System Shifts From Dynamic Fluctuations
![Gillespie simulations-conducted across population sizes of 1,000, 10,000, and 100,000 with initial conditions of [latex]S=0.4[/latex], [latex]I=0.05[/latex], and [latex]R=0.595[/latex]-demonstrate that fluctuations in infectious individuals, when compared to mean-field solutions, exhibit empirically and theoretically derived variances, and that analyzing the quasi-stationary distribution for smaller populations ([latex]N=1,000[/latex]) provides insight into pathogen dynamics as defined in Table 1.](https://arxiv.org/html/2601.14869v1/Figures/high_quasi_sims.png)
New research reveals how inherent properties of changing systems can signal impending transitions, even without a clear tipping point, offering insights for forecasting in fields like epidemiology.
Beyond Scores: Graph Networks Predict Credit Risk with Smarter Connections
A new approach leverages the relationships between borrowers and financial products to improve credit default prediction, moving beyond traditional scoring methods.
When Markets Fall, Emotions Surge: Online Evidence of Loss Aversion
New research analyzing online financial communities confirms that negative emotional responses to market downturns are significantly stronger than positive reactions to booms, reinforcing the psychological principle of loss aversion.
Beyond General Prediction: Testing AI’s Expertise in Real-World Domains
A new benchmark assesses how well AI agents can apply future prediction capabilities to critical sectors like finance, healthcare, and disaster response.
Beyond HAR: Machine Learning Redefines Volatility Prediction
![Forecast accuracy is demonstrably linked to the distribution of out-of-sample volatility, suggesting that predictive models perform best when calibrated to the inherent uncertainty present in dynamic systems-a relationship quantified by [latex] \sigma^2 [/latex].](https://arxiv.org/html/2601.13014v1/x4.png)
New research demonstrates that machine learning models are significantly improving the accuracy of volatility forecasts, challenging established econometric methods.
Decoding the Market with AI: A Knowledge Graph Approach
Researchers are leveraging the power of artificial intelligence and structured knowledge to move beyond traditional stock market analysis and generate explainable investment insights.
Building Agents That Understand the World
New research introduces a framework for aligning artificial intelligence with real-world experience, enabling more robust planning and physical interaction.
When Digital Life Feels Unsafe: Why People Are Reaching Out for Help
A new analysis of millions of online forum posts reveals a surge in concerns about digital privacy, security threats, and the growing need for tailored support.