arrow-backBACK
Healthcare

Figuring Out Applied ML: Building Frameworks and Teams to Operationalize ML at Scale

By February 27, 2020

  • facebook
  • twitter
  • pinterest
  • linkedin

Dr. Nels Lindahl
Director of Clinical Systems
CVS Health

How do you operationalize machine learning at scale? How do you build your team and walk them through the process? Here, Dr. Lindahl shares his strategies for the most productive and money-smart steps toward seamless machine-learning synergy.

Your success may hinge on thinking both inside and outside the box; benefit  from both your homegrown talent and excellent outside consultants to bring a variety of strengths to your machine-learning table:

 

https://youtu.be/snLbMTQ1b5k
Watch Dr. Lindahl's full presentation

Search high and low for your talent 

Expose your existing team to the current “golden age” of machine-learning content; there is a wealth of training content available that may be a perfect fit for your workforce.

Trainings can be done rather inexpensively, but you can only lead a horse to water -- your main challenge is to encourage your staff to take advantage of them.

You can also strengthen your team and get them up to speed quickly by hiring consultants and resourcing information-rich websites like Tensorflow.

Whichever machine-learning toolkits you choose, make sure your team is all on the same page. Working with a number of different kits at the same time could be counterproductive and may not spark synergy. Dr. Lindahl’s best toolkit recommendation: Python.

The more employees you empower and promote within the organization, the more your company will benefit. Teams with a combination of  domain-specific and machine-learning knowledge lend real-world experience to the goals you want to achieve.

“Invest in your folks!” says Dr. Lindahl, and he stresses that this may be the most important takeaway of his presentation. 

Get your talent to work together 

In a perfect world, teams should be self-organizing and all members fall into line and get right to work. However, your inside team itself may need to be introduced to an outside consultant in order to forge a more definitive forward direction. Those consultants who have product knowledge can work effectively with team members who have domain-specific knowledge.

Find both consultants and members of your organization who can best look at your data and know exactly how to give it the ML meaning you need --  and then can make it happen. 

Workflows matter. 

When you’re working with machine learning, you have to figure out how your team and consultants will fit seamlessly into an effective, productive process.

The most difficult task at hand may be attaching to the intensity of content streaming (Dr. Lindahl compares it  to trying to drink water from a fire hose!). A good example of content streaming is self-driving cars, where the content direction toggles between human and non-human.

Whatever the case, your customers and clients will be expecting quick responses, so your models will need to be fast and accurate. When it comes to content streaming, response needs to happen in split seconds. If, for instance, your application programming interface (API) is taking 20-30 seconds to stream, it may become vulnerable or even break.

 

What can create workflow success: high-quality training. 

Adapt a machine-learning strategy.

Once you and your team dig in and create something truly profound and impactful, your next step is to ask yourself: how do I replicate this? 

Find out how to turn your brilliant model into a definable and repeatable process. Make sure it works flawlessly and seamlessly ten, twenty, even thirty times. Search out any component of your system that is transferable -- this can lead to even more interesting possibilities.

Accomplish this and then research how you can employ it in a cost-effective way. 

Know where your ML will reside. 

Be clear about where your machine learning will be. For example, Google, Microsoft and Amazon are already working to make this easier via application programming interface (API). Note that some of your internal systems may not have the hardware to be able to support continued work, use and tuning. When using an API, you want to know the exact use case.

Tell your story: begin with the end in mind. 

Once you and your team have pulled your data set together, where do you want it to go? In order to figure that out, start with the end. Think about how you would tell this story to the executive team, assuring them that the process is definable and repeatable.

Remember to always tie the story to the real world -- for instance, a financial benefit such as increased sales or cost savings. If you’re pitching to an executive team, they’ll be most interested in all things financial.

If you can’t explain how your ML will generate financial benefits, the money people may not fund your next idea. 

 

Recent Posts

https://ai4.io/blog/2020/06/25/how-machine-intelligence-is-saving-lives/
https://ai4.io/blog/2020/06/05/the-top-use-cases-of-ai-in-hospitals/
https://ai4.io/blog/2020/06/05/how-covid-19-is-impacting-the-state-of-ai-in-hospitals/