Hidden Threats in Forecasting AI
A recent competition explored how easily malicious ‘backdoors’ can be concealed within deep learning models used to predict critical time series data.
Science
Cybersecurity’s New Allies: Autonomous Agents for Risk Assessment
A novel multi-agent system, powered by a specialized language model, promises to dramatically accelerate and automate cybersecurity risk evaluations.
Learning to Navigate Growing Networks
![Graph multi-agent reinforcement learning is enhanced through context-aware graph neural networks, enabling agents to leverage relational information and dynamically adjust strategies based on their interconnected environment, effectively optimizing collective performance through informed decision-making within a shared graph structure [latex] G = (V, E) [/latex], where [latex] V [/latex] represents the agents and [latex] E [/latex] their interactions.](https://arxiv.org/html/2603.19501v1/x2.png)
A new framework leverages reinforcement learning to make optimal decisions in dynamic, expanding network environments.
Eyes in the Sky: Smarter Onboard AI for Faster Crisis Response
New research details a distributed AI system that enables Earth observation satellites to analyze data and react to events with unprecedented speed and efficiency.
Beyond the Numbers: Predicting Stock Performance with News and AI
A new approach combines financial time series data with real-time news sentiment analysis to deliver more accurate and generalizable stock price forecasts.
Seizure Control Through Network Harmony
![Optimal control, achieved through Mean Field Game (MFG) techniques, successfully guides the trajectories of key sub-nodes - as evidenced by the shift from uncontrolled red lines to controlled blue lines - and correspondingly reshapes their amplitude probability density functions, bringing them closer to those of a healthy baseline state, a process reflected in the convergence of training losses for both the φ and [latex]G_{\theta}[/latex] functions within the APAC-Net framework.](https://arxiv.org/html/2603.18035v1/Picture/MFG11.png)
Researchers are exploring a new approach to epilepsy treatment by leveraging computational models of whole-brain dynamics to optimize targeted neuromodulation.
Blind Spots in the Code: How AI Review Tools Can Miss Critical Flaws
New research reveals that AI-powered code review systems are vulnerable to confirmation bias, potentially overlooking security vulnerabilities.
Decoding Attacks with AI: A New Era for Security Analysis
Researchers are harnessing the power of artificial intelligence to automatically analyze security incidents, reconstruct attack timelines, and deliver clearer, more actionable intelligence.
Smart Queries: Balancing Power and Cost in Natural Language to SQL Conversion
A new system intelligently combines smaller, efficient AI models with larger language models to translate natural language questions into database queries.
Seeing What Harmful Content Detection Models *Really* See
New research delves into the inner workings of content moderation AI, revealing the complexities of explanation and the critical need for human oversight.