Automated Machine Learning (ML) is an enabler and an accelerator that unleashes the promise for organizations to expedite the analytics life cycle without having data scientists as a bottleneck. In earlier chapters, you learned how to perform automated ML using multiple hyperscalers, including open source tools, AWS, Azure, and GCP.
This chapter, however, is quite different since here we will explore the enterprise use of automated ML. We will look...
Technology decision-makers and stakeholders don't like fads, and you probably shouldn't either. Building and using technology for the sake of technology has limited business value in a vertical enterprise; the technology has to solve a business problem or provide an innovative differentiation to be relevant. Therefore, this inquiry becomes very significant: does an organization really need automated ML or is it just one of those steps in the AI and ML maturity cycle that we can live without? Would this investment result in Return on Investment (ROI), or would it become one of those unused platforms that sounded like a good idea at the time?
Let's try to answer these questions by looking at the value proposition of automated ML and see whether it makes a good fit for your organization. As a technology stakeholder, envision yourself as someone trying to build an enterprise AI playbook and deciding whether to invest in and utilize...
While building your AI playbook and reimagining the AI talent strategy for your organization, you should consider automated ML as an accelerator. The following are some of the reasons why you would want to consider using automated ML for your organization.
Automated ML is rapidly becoming an inherent part of all major ML and deep learning platforms and will play an important part in democratizing advanced analytics. All major platforms tout these capabilities, but for it to be an accelerator for an enterprise, automated ML must play an important role in the democratization of AI. The toolset should enable a citizen data scientist to perform daunting ML tasks with ease and get human-friendly insights. Anything short of explainable, transparent, and repeatable AI and automated ML would not be the advanced analytics accelerator you had hoped for.
We have discussed the benefits of automated ML, but all these advantages are not without their fair share of challenges. Automated ML is not a silver bullet and there are several scenarios where it would not work. The following are some challenges and scenarios where automated ML may not be the best fit.
The size of the dataset is a critical component for automated ML to work well. When feature engineering, hyperparameter optimization, and neural architectural search are used on small datasets, they do not yield good results. The dataset has to be significantly large for automated ML tools to do their job effectively. If this is not the case with your dataset, you might want to try the alternative approach of building models manually.
In a small number of cases, the performance you get from out-of-the-box models may not work – you may have to hand-tune the model for performance or...
Establishing trust in the model trained by automated ML can appear to be a challenging value proposition. Explaining to the business leaders, auditors, and stakeholders responsible for automated decision management that they can trust an algorithm to train and build a model that will be used for a potentially mission-critical system requires that you don't treat it any different from a "man-made" ML model. Model monitoring and observability requirements do not change based on the technique used to build the model. Reproducible model training and quality measurements, such as validating data, component integration, model quality, bias, and fairness, are also required as part of any ML development life cycle.
Let's explore some of the approaches and techniques we can use to build trust in automated ML models and ensure governance measures.
Feature importance...
Now that you have reviewed the automated ML platforms and the open source ecosystem and understand how it works under the hood, wouldn't you like to introduce automated ML in your organization? So, how do you do it? Here are some pointers to guide you through the process.
Andrew Ng is the founder and CEO of Landing AI, the former VP and chief scientist of Baidu, the co-chairman and co-founder of Coursera, the former founder and leader of Google Brain, and an adjunct professor at Stanford University. He has written extensively about AI and ML and his courses are seminal for anyone starting out with ML and deep learning. In his HBR article on AI in the enterprise, he poses five key questions to validate whether an AI project would be successful. We believe that this applies equally well to automated ML projects. The questions you should ask are as follows:
All good things must end, and so does this book. Whew! We covered a lot of ground here. Automated ML is an active area of research and development, and in this book, we tried to give you a breadth-first overview of the fundamentals, key benefits, and platforms. We explained the underlying techniques of automated feature engineering, model and hyperparameter learning, and neural architecture search with examples from open source toolkits and cloud platforms. We covered a detailed walkthrough of three major cloud platforms, namely Microsoft Azure, AWS, and GCP. With the step-by-step walkthroughs, you saw the automated ML feature set offered in each platform by building ML models and trying them out.
The learning journey does not end here. There are several great references provided in the book where you can further do a deep dive to learn more about the topic. Automated ML platforms, especially cloud platforms, are always in flux, so by...
For more information on the topics covered in this chapter, you can refer to the given links:
Did you know that Packt offers eBook versions of every book published, with PDF and ePub files available? You can upgrade to the eBook version at packt.com and as a print book customer, you are entitled to a discount on the eBook copy. Get in touch with us at customercare@packtpub.com for more details.
At www.packt.com, you can also read a collection of free technical articles, sign up for a range of free newsletters, and receive exclusive discounts and offers on Packt books and eBooks.
© 2021 Packt Publishing Limited All Rights Reserved
