No abstract

Adequate knowledge of climatic change over the Tibetan Plateau (TP) with an average elevation of more than 4000 m above sea level (a.s.l.) has been insufficient for a long time owing to the lack of sufficient observational data. In the present study, monthly surface air temperature data were collected from almost every meteorological station on the TP since their establishment. There are 97 stations located above 2000 m a.s.l. on the TP; the longest records at five stations began before the 1930s, but most records date from the mid-1950s. Analyses of the temperature series show that the main portion of the TP has experienced statistically significant warming since the mid-1950s, especially in winter, but the recent warming in the central and eastern TP did not reach the level of the 1940s warm period until the late 1990s. Compared with the Northern Hemisphere and the global average, the warming of the TP occurred early. The linear rates of temperature increase over the TP during the period 1955 -1996 are about 0.16°C/decade for the annual mean and 0.32°C/decade for the winter mean, which exceed those for the Northern Hemisphere and the same latitudinal zone in the same period. Furthermore, there is also a tendency for the warming trend to increase with the elevation in the TP and its surrounding areas. This suggests that the TP is one of the most sensitive areas to respond to global climate change.

We present MT-DNN 1 , an open-source natural language understanding (NLU) toolkit that makes it easy for researchers and developers to train customized deep learning models. Built upon PyTorch and Transformers, MT-DNN is designed to facilitate rapid customization for a broad spectrum of NLU tasks, using a variety of objectives (classification, regression, structured prediction) and text encoders (e.g., RNNs, BERT, RoBERTa, UniLM). A unique feature of MT-DNN is its built-in support for robust and transferable learning using the adversarial multi-task learning paradigm. To enable efficient production deployment, MT-DNN supports multitask knowledge distillation, which can substantially compress a deep neural model without significant performance drop. We demonstrate the effectiveness of MT-DNN on a wide range of NLU applications across general and biomedical domains. The software and pretrained models will be publicly available at https://github.com/namisan/mt-dnn.

No abstract