Researchers have developed a new AI workflow that marries the accuracy of traditional fonts with the creative freedom of image generation to produce remarkably precise and stylish text within images.
A new analysis reveals that while Reversible Instance Normalization can help with distribution shifts in time series data, it doesn’t solve the problem entirely and may even degrade performance under certain conditions.
New research identifies a critical flaw in how AI agents learn to reason actively, and proposes a novel approach to break free from self-imposed limitations.
A new approach enables robots to monitor their actions and identify unexpected deviations during complex tasks, improving reliability in dynamic environments.
![Agent-based modeling reveals how strategically positioned guards ([latex] \text{red patches} [/latex]) can alter elephant movement ([latex] \text{black lines} [/latex]) away from vulnerable human settlements ([latex] \text{yellow regions} [/latex]) and crop-raided agricultural plots ([latex] \text{green squares} [/latex]), with the effectiveness of this intervention differing significantly between a reactive, short-sighted adversary (Myopic Adversary Model) and a more calculating, strategically-minded attacker (Bounded Rationality Stackelberg Attacker Model), both starting from defined locations ([latex] \text{purple circles} [/latex]).](https://arxiv.org/html/2603.11726v1/images/trajectory-summary.png)
New research combines game theory and machine learning to dynamically protect farmland from increasingly resourceful elephant behavior.
![An adaptive navigation system, integrating [latex]GNSS[/latex]/[latex]INS[/latex] data through an ANPMN-UKF framework, allows unmanned ground vehicles to maintain positional accuracy even amidst environmental uncertainty, effectively forecasting and mitigating potential navigational failures.](https://arxiv.org/html/2603.11649v1/figures/Q_R_UKF_method.png)
A new approach combines deep learning with established filtering techniques to significantly improve the accuracy and robustness of unmanned ground vehicle positioning.
![Algorithmic capture-the ability of a model trained on smaller problem instances to generalize to larger ones with minimal additional data-holds for induction and sorting tasks, where the required data scales logarithmically with instance size [latex]C\log(T/T\_{0})[/latex], but breaks down for more complex problems like Shortest Path and Minimal Cut, which exhibit superlinear data requirements even in deep transformer networks, suggesting fundamental limits to generalization based on problem complexity.](https://arxiv.org/html/2603.11161v1/x3.png)
New research reveals fundamental constraints on the computational power of transformer networks, despite their remarkable ability to learn complex algorithms.
![Traffic flow is characterized by a specific terminology encompassing concepts such as [latex] v = \frac{d}{t} [/latex]-velocity, defined as displacement over time-and density, which collectively dictate the dynamics of vehicular movement and underpin strategies for optimizing transportation networks.](https://arxiv.org/html/2603.12096v1/img/traffic_terminology_by_myself.png)
A new multi-agent reinforcement learning framework promises to optimize traffic flow and reduce delays in complex urban environments.
New research explores how incorporating explanations of AI’s reasoning into machine translation models can boost performance and improve the trustworthiness of automated language tools.
New research reveals fundamental limits to training language models on data they’ve already created, highlighting a critical vulnerability known as ‘model collapse’.