concepts without having to stitch together individual container primitives Flexible application modeling supports a wide range of application architectures Small and simple applications are easy, large platform-agnostic application definition for any platform in any environment Consistent application modeling for small devices, Kubernetes on-premises or cloud, and fully- managed cloud environments Extendable business value, not on container primitives and plumbing CRDs combine high-level application modeling with familiar Kubernetes concepts Infra operators continue to use familiar Kubernetes infrastructure

concepts without having to stitch together individual container primitives Flexible application modeling supports a wide range of application architectures Small and simple applications are easy, large platform-agnostic application definition for any platform in any environment. Consistent application modeling for small devices, Kubernetes on prem or cloud, and fully-managed cloud environments. Extendable on business value, not on container primitives and plumbing CRDs combine high-level application modeling with familiar Kubernetes concepts Infra operators continue to use familiar Kubernetes infrastructure

which one is better? GDA makes stronger modeling assumptions, and is more data efficient (i.e., requires less training data to learn “well”) when the modeling assumptions are correct or at least approximately approximately correct, while LR makes weaker assumptions, and is significantly more robust deviations from modeling assumptions. Hence, when the data is indeed non-Gaussian, then in the limit of large datasets, logistic

Which one is better? GDA makes stronger modeling assumptions, and is more data efficient (i.e., requires less training data to learn “well”) when the modeling as- sumptions are correct or at least approximately correct Logistic regression makes weaker assumptions, and is significantly more robust deviations from modeling assumptions When the data is indeed non-Gaussian, then in the limit of large datasets, logistic

config •Network Kubernetes •Core components •Addon •Taint Operations Our thinking of datacenter modeling by extending Kubernetes Onboard Provision Configuration Kubernetes You need onboard something

捕捉时间维度上的短期依赖和维度之间的空间依赖关系 ⚫ Recurrent and Recurrent-skip layer 捕捉长期宏观依赖和周期性信息 ⚫ Autoregresssive 叠加线性比例关系 Modeling Long- and Short-Term Temporal Patterns with Deep Neural Networks Guokun Lai, Wei-Cheng Chang

using PyTorch ○ Huggingface Transformers (transformer models: BERT, GPT, ...) ○ Fairseq (sequence modeling for NLP & speech) ○ ESPnet (speech recognition, translation, synthesis, ...) ○ Most implementations

vkalavri@bu.edu Spring 2020 4/28: Graph Streaming ??? Vasiliki Kalavri | Boston University 2020 Modeling the world as a graph 2 Social networks friend follows The web Actor-movie networks London

74 我們對 Hadoop 的期許： Batch Job Interactive Query Real-Time Processing Graph Processing Iterative Modeling 人心不足蛇吞象 Hadoop 的體質 (Batch Processing) 問題： 每次就是一個 Batch Job，一個接著一個 每個 Batch Job 做的事就是讀入所有資料、處理、寫出結果

heads（12-16个）。这超过了 Transformer论文中的参考配置参数（6个编码器层 ，512个隐藏层单元，和8个注意头） 53 如何训练BERT 方法1：MLM（Masked Language Modeling） 当前词出现不只是单单依靠上文或者下文，其 实应该是同时依赖于上下文深层的双向RNN会 互相透露信息。 句子中有15%的词汇被随机mask掉 交给模型 去预测被mask的部分到底是什么