Taming Shifting Data: A New Approach to Adaptive Filtering
Researchers have developed a novel framework that leverages overparameterization to achieve stable and efficient online learning in dynamic, real-world time series.
Science
Taming Neural Networks: A Guide to Regularization
This review explores the diverse landscape of regularization techniques used to improve the performance and reliability of deep learning models.
Seeing Beyond the Pixel: Hyperspectral Image Resolution Gets a Boost
A new unsupervised learning method uses synthetic data to sharpen the detail in hyperspectral images, offering a path to higher-resolution remote sensing without relying on labeled datasets.
Hijacking AI Finances: A New Threat to Automated Payments
![The Agent Payment Protocol (AP2) [2] establishes a standardized framework for secure and reliable financial transactions between agents.](https://arxiv.org/html/2601.22569v1/Picture1.png)
Researchers have discovered vulnerabilities in Google’s Agent Payments Protocol that could allow malicious actors to siphon funds through cleverly crafted prompts.
Seeing Through the Simulation: How Color Reveals AI-Generated Images
![A discrepancy in color feature map characteristics-specifically, the difference between negative log-likelihood [latex]NLL[/latex] and entropy [latex]H[/latex]-effectively distinguishes photographic images from those generated by AI on the GenImage dataset, as demonstrated by a clear separation in anomaly score distributions computed using DCCT trained solely on photographic content.](https://arxiv.org/html/2601.22778v1/x3.png)
A new technique leverages subtle color patterns inherent in real camera sensors to reliably identify images created by artificial intelligence.
Networked Time: Generating Realistic Data with Complex Systems
![The Inverse Quantile Graph (InvQG) method maps synthetic time series data to quantile graphs, a technique adapted from prior work [Campanharo et al., 2011] and suggesting that complex system behavior can be understood through the distribution of its constituent events.](https://arxiv.org/html/2601.22879v1/inverse_ts_mapping_qg.png)
A new method leverages the principles of network science to create synthetic time series data, offering a compelling alternative to computationally intensive deep learning approaches.
Reasoning Your Way to Fraud Detection
A new approach combines the power of artificial reasoning with graph networks to identify fraudulent activity in complex, text-rich data.
Predicting the Future, One Step at a Time
A new forecasting approach decouples prediction horizons to achieve more stable and accurate long-term time series analysis.
The Search Engine Paradox: Can AI Overviews Undermine Their Own Success?
A new analysis reveals that while AI-powered search summaries initially boost user experience, they risk stifling content creation and, ultimately, search engine revenue.
China’s Past Reveals Climate Secrets, Thanks to AI
A new artificial intelligence framework is unlocking centuries of hidden climate data from historical Chinese records, providing an unprecedented look at past weather patterns.