Search for: All records

This paper proposes and analyzes two new policy learning methods, regularized policy gradient and iterative policy optimization (IPO), for a class of discounted linear-quadratic control (LQC) problems over an infinite time horizon with entropy regularization. Assuming access to the exact policy evaluation, both proposed approaches are proved to converge linearly in finding optimal policies of the regularized LQC. Moreover, the IPO method can achieve a superlinear convergence rate once it enters a local region around the optimal policy. Finally, when the optimal policy for a reinforcement learning (RL) problem with a known environment is appropriately transferred as the initial policy to an RL problem with an unknown environment, the IPO method is shown to converge at a superlinear rate if the two environments are su!ciently close. A model-free version of the policy-based methods is also discussed. Performances of these proposed algorithms are supported by numerical examples. 

Linear Autoencoders (LAEs) have shown strong performance in state-of-the-art recommender systems. However, this success remains largely empirical, with limited theoretical understanding. In this paper, we investigate the generalizability – a theoretical measure of model performance in statistical learning – of multivariate linear regression and LAEs. We first propose a PAC-Bayes bound for multivariate linear regression, extending the earlier bound for single-output linear regression by Shalaeva et al. [45], and establish sufficient conditions for its convergence. We then show that LAEs, when evaluated under a relaxed mean squared error, can be interpreted as constrained multivariate linear regression models on bounded data, to which our bound adapts. Furthermore, we develop theoretical methods to improve the computational efficiency of optimizing the LAE bound, enabling its practical evaluation on large models and real-world datasets. Experimental results demonstrate that our bound is tight and correlates well with practical ranking metrics such as Recall@K and NDCG@K. 

Abstract This study examines the role of human dynamics within Geospatial Artificial Intelligence (GeoAI), highlighting its potential to reshape the geospatial research field. GeoAI, emerging from the confluence of geospatial technologies and artificial intelligence, is revolutionizing our comprehension of human-environmental interactions. This revolution is powered by large-scale models trained on extensive geospatial datasets, employing deep learning to analyze complex geospatial phenomena. Our findings highlight the synergy between human intelligence and AI. Particularly, the humans-as-sensors approach enhances the accuracy of geospatial data analysis by leveraging human-centric AI, while the evolving GeoAI landscape underscores the significance of human–robot interaction and the customization of GeoAI services to meet individual needs. The concept of mixed-experts GeoAI, integrating human expertise with AI, plays a crucial role in conducting sophisticated data analyses, ensuring that human insights remain at the forefront of this field. This paper also tackles ethical issues such as privacy and bias, which are pivotal for the ethical application of GeoAI. By exploring these human-centric considerations, we discuss how the collaborations between humans and AI transform the future of work at the human-technology frontier and redefine the role of AI in geospatial contexts. 

The ocean has absorbed anthropogenic carbon dioxide (C_anthro) from the atmosphere and played an important role in mitigating global warming. However, how much C_anthrois accumulated in coastal oceans and where it comes from have rarely been addressed with observational data. Here, we use a high-quality carbonate dataset (1996–2018) in the U.S. East Coast to address these questions. Our work shows that the offshore slope waters have the highest C_anthroaccumulation changes (ΔC_anthro) consistent with water mass age and properties. From offshore to nearshore, ΔC_anthrodecreases with salinity to near zero in the subsurface, indicating no net increase in the export of C_anthrofrom estuaries and wetlands. Excesses over the conservative mixing baseline also reveal an uptake of C_anthrofrom the atmosphere within the shelf. Our analysis suggests that the continental shelf exports most of its absorbed C_anthrofrom the atmosphere to the open ocean and acts as an essential pathway for global ocean C_anthrostorage and acidification.