Search for: All records

Large language models have achieved remarkable progress in generating human-like text, yet their outputs often lack structural diversity, limiting personalized expression. Recent advances in diffusion models offer new opportunities for text generation, surpassing the limitations of autoregressive paradigms. Building on the idea of hierarchical modeling, we develop a syntax-guided diffusion language model that incorporates latent syntactic variables to enhance text quality, diversity, and personalized conditioning. The proposed framework decomposes the text distribution into a syntactic prior and a conditional text distribution, enabling interpretable control over both syntactic and lexical components. We design a cascaded diffusion process for syntax-to-text generation and extend it to a generalized noncascaded formulation with a unified attention mechanism. To achieve fine-grained personalization, we introduce shared-latent representations that integrate information across users to capture style-specific lexical and syntactic patterns, supporting zero-shot inference. Extensive experiments demonstrate that the proposed approach consistently improves multiple empirical measures of fluency, diversity, and stylistic fidelity. Further qualitative analyses highlight its interpretability and flexibility in learning personalized patterns. 

The Wyoming Craton is often cited as an example of decratonization, implying the removal of its lithospheric keel. However, numerous geophysical imaging studies indicate the presence of thick mantle lithosphere beneath much of the Wyoming Craton. We present a teleseismic P‐wave tomography model produced with data from the Earthscope Transportable Array and two denser temporary arrays. Our model shows high‐velocity anomalies below the crust and to depths exceeding 150 km beneath most of the Wyoming Craton, except the area affected by the Yellowstone Plume. There is good correspondence between our model and previously published seismic and magnetotelluric studies. Based on the geophysical anomalies, the craton can be subdivided into two major blocks located to the NE and SW of the Owl Creek Mountains–Laramie Mountains trend. Each block can be subdivided into two sub‐blocks with roughly N–S oriented boundaries. The eastern boundary of the craton coincides with the Black Hills, which are underlain by a localized low‐velocity anomaly. The external and internal boundaries often have a corresponding topographic expression, and Eocene and younger magmatism occurs either on the periphery of the craton or in the boundary zones between the blocks we identify. Given these findings, we conclude that Laramide deformation utilized zones of weakness in the Wyoming cratonic lithosphere, and may have reduced its thickness, but did not completely destroy it. 

Primary forests play a crucial role in providing essential ecosystem services and supporting biodiversity compared to secondary forests. With increasing threats from extreme climate events and human activities, monitoring primary forest loss is critical for understanding the impact of these threats on ecosystems and biodiversity. Dense time series data from remotely sensed satellite imagery allow us to track historical disturbances, making it an effective source for mapping primary forests over time. However, distinguishing between primary and secondary forests based on spectral-temporal information remains challenging as primary forests can show high resilience to certain natural disturbances (e.g., drought), and secondary forests may not have experienced any disturbance during the satellite observation period. In this context, this study aims to map primary forests on the Caribbean island of Hispaniola using the time series approach and resilience metrics given that primary forests tend to be more resilient than secondary forests. To achieve this, we used spectral-temporal features from COntinuous monitoring of Land Disturbance (COLD) algorithm based on all available Landsat data between 1984 and 2023. Additionally, a resilience map is generated from deseasonalized and detrended spectral observations using the lag-1 autocorrelation method. Then, a Random Forest model was employed to generate an annual primary forest map.