Hub & Spoken.

What's in this podcast?

How to successfully scale machine learning in your organisation 

In this episode Jason talks to Sivan Metzger, Lead of Machine Learning Ops (MLOps) at Data Robot. Jason and Sivan discuss machine learning and why businesses should implement machine learning ops if they want to use machine learning in their business.  

Listen to this episode on Spotify, iTunes, and Stitcher.  You can also catch up on the previous episodes of the Hub & Spoken podcast when you subscribe.

What are your thoughts on this topic? We’d love to hear from you; join the #HubandSpoken discussion and let us know on Twitter and LinkedIn.

One Big Message 

Machine learning should be seen as an investment in a business, not a liability. Over time when implemented properly, machine learning has the potential to deliver an exponential return on your investment.  

[00:30] How Sivan got into Machine Learning and how he discovered a need for MLOps and what his current company Data Robot does 

[06:07] Sivan talks about the four components of MLOps and when you should be implementing machine learning into your business operations  

[13:15] Why AI isn’t implemented as widely as possible and how to gain a level of trust in machine learning 

[19:36] Mindset shifts needed to scale machine learning in companies that are newly implementing it 

[23:00] The biggest challenges most companies face when implementing machine learning 

[25:47] Sivan’s take on the difference between machine learning and AI 

[26:30] How machine learning can be democratised so it’s not just reserved for those in ‘ivory towers’ 

What is machine learning operations? 

MLOps is, at its core, about getting different people with different skills to work together on machine learning solutions for the company. Even though the people you need to implement ML in a business may be in different departments, it is still critical that they frequently communicate with each other when coming up with ML solutions. While the role of MLOps is a play on ‘DevOps,’ both roles still have a lot in common at face value: both need to facilitate easy collaboration to help accelerate delivery.  

MLOps also combines machine learning with DevOps principles by applying ML models to operational tasks. This can include monitoring production systems for anomalies or automating human processes like infrastructure provisioning or hardware maintenance.   

MLOps can be used to automate time-consuming manual tasks that are tedious for humans – but easy for machines – and provide real-time insights into system performance before they become a problem.  

Four components of machine learning operations (MLOps) 

When you are trying to implement machine learning, there are four components that will help you systemise for success.  

1. Deployment: Make sure everything is in a pipeline and that the deployment is as streamlined as possible.  
2. Monitoring: A business must have insights into ML. Because of its sometimes unpredictable nature you need to specifically monitor for it.  
3. Life cycle: The life cycle in ML pertains to everything that surrounds the deployment. For example, you need to check how data is gathered and who is responsible for correcting any problems. It is the culmination of different people with different skill sets coming together to create a system so that everything runs smoothly.  
4. Model governance: you need to monitor production model governance surrounding how machine learning occurs – what information is important and what material it is given to learn.  

Why AI can’t replace humans entirely 

AI isn’t adopted as widely today as it could be due to people. This is a good thing because it is never wise to completely go all in on new technology, especially when you are not sure what the outcome might be. While machine learning can instill fear into people over job security or unpredictability over new technology, we may never get to the point where AI and technology cover all their potential domains.  

In addition to this, machine learning does require human intervention for error identification and compliance. In the finance industry if your technology isn’t compliant your institution faces the possibility of being shut down. We still need humans to be able to monitor and critically analyse the behaviour and output of machine learning until trust is built or a re-training model is deployed.  

Implementing machine learning in a business 

Machine learning is a field of computer science that has seen a huge increase in use and development over the past decade, which means that older businesses can sometimes find it very hard to adopt this new technology.  

It is not always affordable or easy for an older company with legacy systems, but it can be done. When trying to implement ML into an older business you need to be mindful that ML is an investment when done correctly has the potential to give a return on itself multiple times over.  

To summarise 

 Machine learning is a rapidly changing field, with businesses adopting it into their business operations as it becomes more accessible. Because of this, it is important for businesses who use ML to invest in MLOps. It will help to streamline your processes, improve deliverability and ensure that everything is running smoothly.