Can AI Agents Survive the Markets?
A new benchmark reveals that artificial intelligence models designed for financial trading often prioritize textbook knowledge over practical resilience in volatile conditions.
Science
Evolving How We Search: EvoX Automates Algorithm Discovery
A new framework, EvoX, dynamically adapts search strategies during optimization, achieving consistent performance gains across diverse problem types.
Steering Offline RL Towards Robust Policies
![Under a no-shift setting, the Dynamic Relative Policy Update (DRPU) demonstrably drives cumulative error - measured as [latex] err_k [/latex] at iteration 80 - towards zero, achieving convergence to a comparator policy [latex] \pi_{cp} [/latex], while the Local Search Policy Update (LSPU) plateaus at a suboptimal policy and incurs a persistent, non-vanishing error, highlighting a fundamental difference in their capacity to navigate policy space.](https://arxiv.org/html/2602.23811v1/2602.23811v1/x1.png)
New algorithms address key challenges in offline reinforcement learning, offering improved performance and theoretical guarantees for policy optimization.
Beyond Keywords: How AI-Generated Data is Reshaping E-commerce Search
A new approach leverages the power of artificial intelligence to understand and respond to complex, niche product searches in online retail.
Smarter IoT Security: Federated Learning Detects Anomalies Across Diverse Networks
A new approach to federated learning enables robust anomaly detection in Internet of Things networks, even when devices produce vastly different data.
Spotting the Fake: A New Way to Detect AI-Generated Images
![The capacity to distinguish between real and synthetically generated images hinges on the generator's proficiency; a weak generator produces images with distributions markedly different from real images - resulting in larger reconstruction errors [latex]\Delta_{\textrm{fake}}(y)[/latex] compared to those of real images [latex]\Delta_{\textrm{real}}(x)[/latex] - while a strong generator collapses this distinction, rendering synthetic images difficult to identify due to comparable reconstruction errors and a projection [latex]\Pi_{\mathcal{M}}[/latex] onto the manifold [latex]\mathcal{M}[/latex] that closely mirrors real image distributions.](https://arxiv.org/html/2602.23732v1/2602.23732v1/figures/illustration.png)
As AI image generation becomes increasingly sophisticated, researchers are developing methods to reliably distinguish between real and synthetic content.
Hunting Gravitational Waves: A New Tool for Eliminating Detector Noise
Researchers have developed a machine learning system to automatically pinpoint the sources of transient noise in gravitational wave detectors, boosting the chances of spotting faint signals from the cosmos.
Beyond the Algorithm: Finding Your Next Favorite Movie
![Recommender algorithms exhibit varied performance characteristics, with those leveraging collaborative filtering demonstrating superior accuracy-quantified by [latex] RMSE [/latex]-compared to content-based approaches, though hybrid methods incorporating both achieve a balanced optimization of precision and recall.](https://arxiv.org/html/2602.24125v1/2602.24125v1/MAPE.png)
A new study dives into the effectiveness of various machine learning techniques for building accurate and personalized movie recommendation systems.
Seeing is Understanding: A New Approach to Document AI
Researchers have developed a system that intelligently extracts only the necessary information from visual documents, dramatically improving performance in question-answering and information retrieval.
Shopping Smarter: AI Agents Trained with Synthetic Data Boost E-Commerce Research
A new framework uses simulated shopping experiences to train artificial intelligence agents to conduct more effective product research online.