The Challenges and Benefits of AutoML

Machine Learning and Artificial Intelligence have revolutionized how organizations are utilizing their data. AutoML or Automatic Machine Learning automates and improves the end-to-end data science process. This includes everything from cleaning the data, engineering features, tuning the model, explaining the model, and deploying it into production. AutoML accelerates your AI initiatives and can help make data scientists more effective and efficient at solving problems and providing business value. 

Last week, we hosted a live broadcast on H2O.ai’s Linkedin with John Spooner, EMEA Head of Artificial Intelligence at H2O.ai, and Harib Bakhshi, Lead Data Scientist at H2O.ai, who discussed the Challenges and Benefits of AutoML. 

Check out some key takeaways of this conversation:

• What is AutoML?

For John and Harib, there tends to be a lot of ambiguity in the terminology that gets used in the industry when people talk about AutoML. If we consider the Machine Learning process and everything it takes to build the Machine Learning model, AutoML consists of trying to automate as much of that process as possible. AutoML is looking at areas where there is repeated work, it analyzes how certain parts of that process can be enhanced and makes data science teams more effective by automating all these processes.

• Why is there such a big interest in Machine Learning now?

There is a lot of excitement around Machine Learning and AI and Harib sees several reasons for that. First, it is important to note that the math that gets used in the backend is old, this hasn’t changed over the years. What has changed though, is that there has been a serious reduction in compute costs over the past few years, so it is now cheaper to have very powerful computers to process the calculations. Another reason to explain the excitement around AI and Machine Learning is that these technologies allow us to handle different kinds of datasets: numeric, text, image… which considerably expands the range of possibilities. 

• Accountability is still key

The downside of having Machine Learning models making automated decisions on a daily basis is that it makes it difficult to determine who is accountable for these decisions. In regulated industries like Financial Services or Healthcare, accountability is key. Harib recommends having some manual checkpoints where human beings can intervene and sign off on parts of the process. This can add the accountability needed and help with the regulation and governance.  

H2O Driverless AI makes it possible to automate the machine learning processes and helps the data science teams to work on projects faster and more efficiently. Find out more here.

Interested in watching the recording of the session and learning more? Click here.