Hi, welcome to this specialization on Machine Learning Engineering for production, MLOps. Say you've trained an accurate machine learning model in the Jupiter Notebook, you celebrate that and you might now be wondering, now what do I do. After reaching the milestone of developing a good model, actually putting your model into production. They have a working system that is making useful predictions. That's still requires much more. This specialization will teach you the skills you need to build and deploy production machine learning systems. You learned about the full machine learning process lifecycle from scoping out your project, to data, to modeling through deployment. You're better execute the machine learning project all the way from start to finish. One common misconception is that the only difference between developing model that you fit the notebook on your local machine and deploy that model into production is literally just the deployment piece, maybe just a bunch of software engineering. But that's not true, it's not just a bunch of software engineering. There are machine learning challenges to production as well that on edges in the process of developing a model into Jupiter notebook. This discipline of building and maintaining production systems, that is the processes in tools for doing all this is sometimes called MLOps or machine learning operations. You'll learn about that too. I'm thrilled to be bringing you this new specialization with two fantastic instructors from the google TensorFlow team, Robert Crowe and Laurence Moroney. Robert Crowe is it TensorFlow developer, Engineer at Google, a Data Scientists and TensorFlow advocate. Robert has a passion for helping developers quickly learn what they need, hopefully what you need to be productive. Lawrence Moroney, these AI advocacy at google and is also instructor of three other TensorFlow specializations with deep learning AI and is also the book, AI and machine learning, if we call this. Glad to have both of you with us for the specialization. I'm delighted to work with you again, it's always a pleasure. As you said Andrew, in addition to building a first working model, in a production system, you need to handle a whole range of issues, including things like data drift, where the distribution of the data you trained on maybe eventually become very different from the distribution of the data that you're running inference on. A key topic that we'll discuss is change. The world changes and your model needs to be aware of that change. In this specialization, we'll also introduce you to several related topics outside of machine learning. You can look at Production Machine Learning as both machine learning itself and the knowledge and skills required in modern software development. If you're working on a machine Learning Team and industry, you really need expertise in both machine learning and software to be successful. This is because your team will not just be producing a single result. You'll be developing a product or service that will operate continuously and maybe a mission critical part of your company's work. That's right. I can say from my own experience that oftentimes the most challenging aspects of building machine learning systems turn out to be the things that you least expected like deployment. It's all very well being able to build a model, but getting that into people's hands and seeing how they use it can be very eye-opening. You might think you have the perfect model for the perfect scenario, but your users could have different opinions and it's always really good to learn from them. They might be, for example, okay, with a round robin trip to a server for a frequently updated model. Or they might insist that their data never leaves their device. So you need to know the best ways to keep their on-device models really fresh. This specialization contains four causes that provide you with the knowledge and hands-on practice to get a Production Machine Learning system up and running. In the first course which I teach, you see an overview of the entire life cycle of a production machine learning project, from the scoping to getting data, to modeling, to deployment. The second course is all about data and how it evolves over time. With this course, you'll build data pipelines by gathering, cleaning, and validating data sets using TensorFlow extended or TFX and the TFX family of Libraries. To understand your data evolution, you'll track changes with ML metadata using data provenance as a conceptual framework. The third course is focused on machine learning modeling pipelines in production. In this course, you'll learn how to manage modeling resources to best serve inference requests and minimize costs. You'll also use analytics to address model fairness, explain-ability issues, and mitigate bottlenecks. Course 4 is all about deployment. This means of course that you want to get ready to serve your users requests. Now this can be exciting and challenging all at the same time. In course 4, you'll deliver deployment pipelines from model serving that might require many different infrastructures. You'll also apply best practices to maintain a continuously operating production system that stays up to date and importantly keeps your users needs front of mind. As a learner in this specialization, we're assuming that you're comfortable with Python programming and Machine Learning and that you have some familiarity with one of the deep-learning frameworks in Python, like TensorFlow, keras, or PyTorch. If you've completed the deep learning specialization offered by deep-learning data AI, you should be in great shape to start this specialization. Alternatively, of course, if you've completed the deep learning data AI TensorFlow developer certificate program, you'll be really well-prepared to start the specialization. Great. Well with that, let's get started. That sounds great to me. Sounds good. Let's do it.
