Search for: All records

Abstract Uttar Pradesh, with a population of 237 million, is the largest agrarian state in India, located in the Indo‐Gangetic plains. Rice cultivation is widespread across all districts of Uttar Pradesh, which have varying climate regimes, irrigation infrastructures, crop management practices, and farm sizes. The state is characterized by different agroecological zones (AEZs) with semi‐arid to sub‐humid climates with significant variability in monsoon rainfall. In this study, the impact of climate change on Kharif‐season rice is estimated using crop‐climate scenarios in Uttar Pradesh. A process‐based Crop Simulation Model, Crop Estimation through Resource and Environment Synthesis‐Rice, was simulated with bias‐corrected and downscaled climate data for historical (1995–2014) and three future periods (the 2030s, 2050s, and 2090s) for two mitigation pathways (SSP2‐4.5 and SSP5‐8.5) from the Coupled Model Intercomparison Project 6. Phenology, irrigation amount, crop evapotranspiration, yield, and water use efficiency were evaluated and assessed for all AEZs. Based on the ensemble of 16 climate models, rainfed rice yield increased in the AEZs of western Uttar Pradesh due to increased rainfall, while in eastern Uttar Pradesh yield decreased, under both shared socioeconomic pathways (SSPs). Irrigated rice yield decreased in all AEZs under both SSPs due to an increase in temperature and a decrease in the length of the growing period, with reductions of up to 20% by the 2090s. Irrigation requirements decreased from the 2030s to the 2090s due to increased rainfall and decreased crop evapotranspiration. Despite the projected increase in rainfed yield, the overall rice yield is expected to decrease in the future under both SSPs. 

Knowledge distillation aims at reducing model size without compromising much performance. Recent work has applied it to large vision-language (VL) Transformers, and has shown that attention maps in the multi-head attention modules of vision-language Transformers contain extensive intra-modal and cross-modal co-reference relations to be distilled. The standard approach is to apply a one-to-one attention map distillation loss, i.e. the Teacher’s first attention head instructs the Student’s first head, the second teaches the second, and so forth, but this only works when the numbers of attention heads in the Teacher and Student are the same. To remove this constraint, we propose a new Attention Map Alignment Distillation (AMAD) method for Transformers with multi-head attention, which works for a Teacher and a Student with different numbers of attention heads. Specifically, we soft-align different heads in Teacher and Student attention maps using a cosine similarity weighting. The Teacher head contributes more to the Student heads for which it has a higher similarity weight. Each Teacher head contributes to all the Student heads by minimizing the divergence between the attention activation distributions for the soft-aligned heads. No head is left behind. This distillation approach operates like cross-attention. We experiment on distilling VL-T5 and BLIP, and apply AMAD loss on their T5, BERT, and ViT sub-modules. We show, under vision-language setting, that AMAD outperforms conventional distillation methods on VQA-2.0, COCO captioning, and Multi30K translation datasets. We further show that even without VL pre-training, the distilled VL-T5 models outperform corresponding VL pre-trained VL-T5 models that are further fine-tuned by ground-truth signals, and that fine-tuning distillation can also compensate to some degree for the absence of VL pre-training for BLIP models. 

An interspersed array of Cs and Rb atoms was used to implement a protocol for the correction of correlated errors. 

Smart home devices transmit highly sensitive usage information to servers owned by vendors or third-parties as part of their core functionality. Hence, it is necessary to provide users with the context in which their device data is collected and shared, to enable them to weigh the benefits of deploying smart home technology against the resulting loss of privacy. As privacy policies are generally expected to precisely convey this information, we perform a systematic and data-driven analysis of the current state of smart home privacy policies, with a particular focus on three key questions: (1) how hard privacy policies are for consumers to obtain, (2) how existing policies describe the collection and sharing of device data, and (3) how accurate these descriptions are when compared to information derived from alternate sources. Our analysis of 596 smart home vendors, affecting 2, 442 smart home devices yields 17 findings that impact millions of users, demonstrate gaps in existing smart home privacy policies, as well as challenges and opportunities for automated analysis. 

The success of our project depended on forming a trusting, collaborative relationship with teachers and conducting in-depth interviews with adolescents who had never met our team. Yet, when we were ready to launch this work, it was not safe to meet with teachers or students in-person due to the COVID-19 pandemic. Nor was it possible to conduct site visits to the school. How could we continue collaborative module development and conduct meaningful research when the in-person methods we planned for our study were no longer feasible given the health and safety challenges of the pandemic? Our team had to make important decisions about module development and deployment while keeping students' and teachers' health and safety in mind. Rather than focusing on the problems of being unable to perform face-to-face data collection or module development, we began exploring new and alternative technological solutions. Building trust and engaging teachers while negotiating the communicative and relational restrictions inherent in online interactions was another challenge addressed in the paper. 

Abstract Aging, often considered a result of random cellular damage, can be accurately estimated using DNA methylation profiles, the foundation of pan-tissue epigenetic clocks. Here, we demonstrate the development of universal pan-mammalian clocks, using 11,754 methylation arrays from our Mammalian Methylation Consortium, which encompass 59 tissue types across 185 mammalian species. These predictive models estimate mammalian tissue age with high accuracy (r > 0.96). Age deviations correlate with human mortality risk, mouse somatotropic axis mutations and caloric restriction. We identified specific cytosines with methylation levels that change with age across numerous species. These sites, highly enriched in polycomb repressive complex 2-binding locations, are near genes implicated in mammalian development, cancer, obesity and longevity. Our findings offer new evidence suggesting that aging is evolutionarily conserved and intertwined with developmental processes across all mammals.