azure – Conversational AI on Microsoft Platform
During the pandemic, we all found ourselves in isolation, and relying more and more on effective electronic means of communication. The amount of digital conversations increased dramatically, and we need to rely on bots to help us handle some of those conversations. In this blog post, I give brief overview of conversational AI on Microsoft platform and show you how to build a simple educational bot to help students learn. Read More ›
Azure – Azure
Unlike theoretical data science, applied data science involves additional steps to manage the lifetime of a model, which are commonly called MLOps. Azure Machine Learning is a service that conveniently supports MLOps practices, so getting to know it seems like a good idea. In this post, I will talk about the best way to learn (Udacity), and to demonstrate your knowledge (Certification). Read More ›
azure – Azure
It is estimated that a Data Scientist spends about 80% of her time on data preparation, and not on the model training. If your training setup is just a GPU virtual machine in the cloud - it means that you are spending 80% of its uptime in vain, because GPU is not utilized. For more cost-effective way we may want to split the work between two virtual machines, one for data preparation, and another one for actual training. This is exactly the setup I used for some time when working in Microsoft CSE, before switching to AzureML, so I will share my knowledge here. Read More ›
datascience – Machine Learning
В феврале 2020 года команда центра цифрового развития АСИ позвала меня войти в состав экспертов международного конкурса World AI & Data Challenge. В этой заметке я немного расскажу о самом конкурсе, а также о том, как можно начать решать одну из интересных задач этого конкурса - распознавание шрифта Брайля. Read More ›
science – Science
One of the main characteristics of an epidemic is the effective reproduction number (Rt), which indicates the number of people each infected individual will further infect at any given time. Being able to estimate Rt is an important task, because this number defines whether the epidemic is expected to grow (Rt>1), or will start declining (Rt<1). In this post, I suggest sliding SIR method of estimation of Rt based on fitting SIR epidemic model to the infections data in different countries. Read More ›