Amazon Web Service Japan Co., Ltd. is Obuchi of Machine Learning Solution Architect.The AWS JAPAN Meguro Office regularly holds the "Amazon Sagemaker Case Festival" (Twitter: #sageMaker_fes).At the 7th AMZON SAGEMAKER Case Festival held on August 29, 2019, the latest information and technical information of services by AWS JAPAN's solution architect and customers using Amazon Sagemaker are invited to guest speakers.He talked about the "experience story" by the customer who actually introduced it.
Amazon Web Service Japan Co., Ltd. Machine Learning Solution Architect Masaki Samejima
The whole image of the machine learning service provided by AWS was introduced based on the AWS concept, called Machine Learning Service Stack. The Machine Learning Service Stack is divided into three layers of AWS machine learning services: AI services, ML services, ML frameworks and infrastructure. Among them, we introduced AI services that allow customers to use machine learning from the API just by preparing data. You can use a variety of machine learning algorithms, such as Amazon Rekognition, which recognizes still images and video recognition, and Amazon Personalize, which uses the same thing as the technology used in Amazon. Finally, we introduced AI Services Amazon Forecast, which will predict the time series data that became GA on August 23, with demonstrations. Amazon Forecast can be used not only from the AWS console but also from the Jupyter notebook on Amazon Sagemaker via API.
Amazon Web Service Japan Co., Ltd. Machine Learning Solution Architect Seiko Utsunomiya
We talked about Amazon Sagemaker, a managed service for machine learning provided by AWS. As of August 2019, Amazon Sagemaker can be used in 17 regions including Tokyo Region. In the development of a machine learning system, Amazon Sagemaker reduces the burden on environment construction so that customers can focus on parts that can differentiate business, such as model development. By using a pre -prepared container image for Amazon ECR, you can launch a Jupyter environment in a few minutes, and you can execute a learning job in seconds. There are various functions, such as Elastic Inference and batch inference, which can easily host endpoints with APIs and reduce inference environments for cost reduction wards. Amazon Sagemaker has released a continuous new feature. In this announcement, we introduced that spot instances can be used during learning and that Amazon EFS and Amazon FSX for LUSTRE can now be used for data downloading. In addition, he introduced new features such as his R compatible Jupyter notebook, random search for hyper parameters, and search functions for training jobs. Many functions are improving based on feedback from customers, so please let us know your voice.
Amazon Web Service Japan Co., Ltd. Solution Architect Kamihara
Teachers who use their own data need to prepare labeled data to learn. Anotation (labeling to data) takes costs and time, so today it is often one of the major hurdles of the use of machine learning. Amazon Sagemaker Ground Truth provides functions for annotation. Using Amazon Sagemaker Ground Truth, labels to images, images, object detections in images, semantic segmentation that labels in pixels, text classification, easy annotations using the GUI. It will be realized. In August 2019, a template for labeling unique expressions (human name, place name, date, etc.) was added to the text, and all of his five built -in labeling tools were available. In addition, you can choose from three types of annotation workers: public sourcing (Amazon Mechanical Turk), a private, registered and used friends and employees, and a vendor requested by registered annotation companies. can do. In order to increase the accuracy of annotation, we have introduced how to integrate multiple people labeling one data to integrate them. The labeling data created using Amazon Sagemaker Ground Truth can be used as it is on Amazon Sagemaker.
CAC Crois Fharamaco Viganth Specialist: Takeshi Takagi System Architect/Machine Learning Engineer: Hideki Inoue
CAC Crois Co., Ltd. is working on managing safety information about pharmaceuticals (collecting and analyzing side effects). Until now, we have provided the process of registering information in one example database, evaluating whether or not side effects, and reporting as a side effect if it is side effects. The purpose is to improve this work efficiently. CAC Crois Co., Ltd. is currently using 66 AWS services and introduced initiatives using Amazon Sagemaker, one of them. In analyzing the side effects of pharmaceuticals, it was necessary to first unify the collected information into a term used in a dictionary for medical products called Meddra (for example, the expression "body" is "fatigue". Replacement, etc.). The terms of the collected information were treated as a classification problem that corresponds to the medra terms, and the classification model for that was developed using Amazon Sagemaker. In order to learn a huge classification model for medra terms, we have verified multiple natural language algorithms. When using Amazon Sagemaker, we introduced the region selection to save costs, how to execute it from local, and how to deploy multiple models.
Suntory System Technology Co., Ltd. Technology Department / Specialist: Motosei Takagi
The Advanced Technology Department of Suntory System Technology Co., Ltd. is working on the development of a system that can support life science with the latest information science and develop new businesses. As a result of the Advanced Technology Department, he introduced genetic data analysis, etc., and talked about the development of document classification system using Amazon Sagemaker as a case of machine learning. As for the document classification, he has not been able to produce the expected performance while being examined by the deep learning method such as his machine learning and CNN, and he is now implemented using BERT. By using Amazon Sagemaker, he was able to immediately use his POC to use it on site thanks to a full -managed environment. As a change required to use your own machine learning algorithm in Amazon Sagemaker, you can explain in detail how to change environment variables and how to pass HPO, and there are not many changes. He told me that he was able to move. In addition, when incorporating the tuning of hyper parameters using Optuna, we have implemented the AWS blog. Finally, what has changed using Amazon Sagemaker, being able to advance the POC with a partner quickly, that overseas engineers can participate in the development without being bound by the GPU server in the office, security settings and user management, etc. He told me that the operation work was simplified.
This time, we introduced Amazon Sagemaker and Annotation Service Amazon Sagemaker Ground Truth from the entire AWS machine learning service, and welcomed guests who are using Amazon Sagemaker and talked about actual examples.Next time, the 8th Amazon Sagemaker Case Festival | Experience Hands -on is scheduled for September 19, 2019.You can see the overview of the past Amazon Sagemaker Case Festival and the stage slide from the link below.