加利福尼亚大学洛杉矶分校

Adapting Speech Foundation Models (SFMs) to intersectional domains remains challenging when multiple distribution shifts such as acoustic conditions, speaking style, or speaker population co-occur and joint supervision data is scarce. Rather than training a separate model for each domain combination, this work studies compositional domain adaptation, asking whether model updates induced by distinct supervision sources can be recombined to generalize to unseen intersections. We investigate this question using child automatic speech recognition (ASR) as a case study, where age-related acoustics and speaking style jointly define a difficult low-resource setting. We propose a structured model merging framework that composes task-specific adaptations without retraining. To enable stable and interpretable composition, we introduce Headwise Selective Attention (HSA) Merge, which restricts parameter arithmetic to salient attention heads where task-specific adaptations are concentrated. We further demonstrate robustness under acoustic mismatch, cross-corpus transfer, dialectal variation, and when source vectors are derived from synthetic or noisy data. Across Whisper model sizes, HSA Merge consistently outperforms single-source fine-tuning by relative WER of up to 20.6% on child ASR and 29.2% on cross-dialect transfer. These results indicate that key factors of variation in ASR models behave as partially separable components and that structured attention-level merging provides a practical mechanism for compositional generalization in low-resource and intersectional ASR settings.

Purpose: HER2 amplification occurs in 18% to 27% of gastric and gastroesophageal junction cancers. Lapatinib, a potent ATP-competitive inhibitor simultaneously inhibits both EGFR and HER2. To explore the role of HER family biology in upper gastrointestinal cancers, we evaluated the effect of lapatinib, erlotinib, and trastuzumab in a panel of molecularly characterized human upper gastrointestinal cancer cell lines and xenografts. Experimental Design: EGFR and HER2 protein expression were determined in a panel of 14 human upper gastrointestinal cancer cell lines and HER2 status was assessed by fluorescent in situ hybridization. Dose-response curves were generated to determine sensitivity to lapatinib, erlotinib, and trastuzumab. In HER2-amplified cells, the combination of trastuzumab and lapatinib was evaluated using the median effects principal. The efficacy of lapatinib, trastuzumab, or the combination was examined in HER2-amplified xenograft models. Results: Lapatinib had concentration-dependent antiproliferative activity across the panel with the greatest effects in HER2-amplified cells. There was no association between EGFR protein expression and sensitivity to any of the HER-targeted agents. Cell cycle analysis revealed that lapatinib induced G1 arrest in sensitive lines and phosphorylated AKT and phosphorylated ERK were decreased in response to lapatinib as well. The combination of lapatinib and trastuzumab was highly synergistic in inhibiting cell growth with a combination index of <1. The combination also induced greater decreases in AKT and ERK activation, G0-G1 cell cycle arrest, and increased rates of apoptosis. In vivo studies showed that the combination of lapatinib and trastuzumab had greater antitumor efficacy than either drug alone. Conclusion: Together, these data suggest that lapatinib has activity in HER2-amplified upper gastrointestinal cancer and supports the ongoing clinical investigation of lapatinib in patients with HER2-amplified disease. Clin Cancer Res; 16(5); 1509–19

Africa is increasingly being exposed to the negative impacts of climate and environmental change, while having less capacity to respond compared to other continents. The vulnerability partially results from unprecedented demographic growth, urbanization, and industrialization. However, the continent has still largely been underserved by the broader Earth system science (ESS) community, as evidenced by the limited amount of ESS data and research that cover Africa compared to other areas of the world. Here, we present the recent University Corporation for Atmospheric Research (UCAR) Africa Initiative that aims to enhance environmental sustainability in Africa by fostering international collaborative research partnerships coled by African scientists. Specifically, we outline urgent challenges and opportunities identified through an international workshop in six areas of ESS, namely, 1) air quality and health, 2) weather, 3) climate, 4) land and water, 5) social science perspectives, and 6) developing equitable collaboration and sustainable infrastructure. We highlight examples of successful partnerships and conclude with recommendations to advance collaborative, actionable ESS research that addresses Africa's critical environmental challenges.

This study presents a new physical-biogeochemical simulation of the Pacific Ocean that resolves mesoscale dynamics and explicitly includes tidal forcing. The primary objective is to develop and document a modeling framework that serves both as a detailed record of model configuration and forcing preparation, and as a reference for future regional downscaling. The model is extensively evaluated against physical and biogeochemical data sets and successfully reproduces large-scale circulation patterns and key biogeochemical features. The second objective is to assess the impact of explicitly including tidal forcing on primary production and carbon export, thereby clarifying the biogeochemical consequences of tidal dynamics. While tides are known to energize high-frequency motions, their influence on ocean biogeochemistry remains insufficiently constrained. By comparing simulations with and without tidal forcing, we show that tides enhance net primary production and particulate organic carbon export, particularly in coastal zones and the eastern equatorial Pacific. Basin-wide, tidal forcing increases biological carbon uptake by approximately 0.45 PgC and export by 0.09 PgC . This enhancement is primarily driven by tide-induced advection-mixing interactions, which increase nutrient supply to the surface and stimulate biological productivity. These findings underscore the importance of including tides in ocean biogeochemical models to improve estimates of the global carbon cycle and refine projections of future oceanic carbon uptake.

Analyzing the dynamics of trapped electron fluxes in the Earth's outer radiation belt is a complex task, due to the presence of insufficiently known parameters and the long runtimes of multi-dimensional radiation belt codes, preventing a thorough examination of dependencies on all parameters. Here, we present an approximate eigenfunction modeling of whistler-mode wave-driven electron pitch-angle diffusion, slightly generalized compared to previous work. This new model can approximately describe, in an easy, flexible, and fast way, both the asymptotic electron pitch-angle distribution (PAD) at all pitch angles and its temporal evolution toward this final state, in both weak and strong diffusion regimes, in the presence of a finite, time-varying electron source. In this model, wave-driven pitch-angle diffusion is assumed to prevail over energy diffusion and radial diffusion, limiting its applicability to the plasmasphere or intervals of smooth decay of the electron flux outside the plasmasphere, during moderately active periods. We propose a new method, based on this model, for estimating the energy spectrum and temporal variation of the electron source. We investigate the dynamics of the electron flux measured by the Van Allen Probes and Arase spacecraft during two events in 2018 and 2022 in the outer radiation belt. We demonstrate that the new model can reproduce the evolution of the measured electron flux and of its PAD, provided that the magnitude of diffusion rates is normalized to the observed decay timescale in the 300-600 keV range and that a finite electron source term is included below 300 keV.