Intelligent Sensing at the Edge: A New Era for Wireless Networks
This review explores how collaborative artificial intelligence is enabling advanced distributed sensing capabilities in next-generation wireless systems.
Science
Unlocking AI’s Weaknesses: A Deep Dive into Large Language Model Security
![A multi-layered defensive framework sequentially refines threat assessment through pattern screening, semantic understanding, behavioral categorization, and active learning, achieving progressively deeper analysis while prioritizing minimal latency for real-time deployment-a design predicated on the principle that [latex] \text{Accuracy} = f(\text{Depth}, \text{Latency}) [/latex].](https://arxiv.org/html/2603.17123v1/defensive_framework_diagram-3b2.png)
New research exposes critical vulnerabilities in leading large language models and introduces a robust framework for detecting and mitigating potential attacks.
Tracking Stories as They Shift: Monitoring Narratives in Real Time
A new approach allows for the continuous tracking of evolving narratives within fast-moving information streams, like social media, by focusing on semantic changes rather than fixed topics.
Harnessing Chaos: Neural Networks Powered by Unpredictability
[/latex] - suggesting that complex information processing can arise from simple, interconnected systems driven by internal dynamics.](https://arxiv.org/html/2603.16909v1/x3.png)
Researchers are exploring how networks of chaotic oscillators can be trained using machine learning techniques to achieve robust pattern recognition and signal processing.
Predicting Solver Success with Early AI Insights
![The proposed parallel scheme demonstrates rapid convergence across varied initialization scenarios, as evidenced by the decreasing order of magnitude of error [latex] \log_{10}(E^{(k)}) [/latex] with each iteration.](https://arxiv.org/html/2603.16980v1/E1c.png)
New research demonstrates how machine learning can rapidly assess the reliability of root-finding algorithms, drastically reducing computational overhead.
Predicting Power Grid Behavior with AI and Physics
![The system architecture leverages a four-layer, residual Graph Attention Network (GATv2) encoder-informed by bus-type awareness and a supervision mask-to predict voltage magnitudes [latex]V_m[/latex], phase angles δ, and both active [latex]P_g[/latex] and reactive [latex]Q_g[/latex] power generation, all within a unified decoding trunk and guided by a physics-informed loss function [latex]\mathcal{L}_{phy}[/latex] that incorporates predicted outputs and supervisory signals.](https://arxiv.org/html/2603.16879v1/x1.png)
A new framework leverages graph neural networks and physics-informed learning to accurately and adaptively model complex power system flows.
Turning the Tide: Real-Time Flood Forecasting with Community Data
A new framework leverages the power of crowdsourced observations and physics-based modeling to dramatically improve the accuracy of urban flood impact predictions.
AI-Powered ECG Analysis: A New Era for Digital Heart Health
A comprehensive new system leverages deep learning to deliver accurate arrhythmia detection and scalable ECG analysis for improved patient care.
Decoding Earth’s Signals with Smarter Machines
A new review explores how machine learning, grounded in physics, is transforming our ability to interpret seismic and volcanic activity.
Building AI Economies That Last
A new architecture prioritizes verifiable robustness and safety as core principles for designing AI-driven economic systems.