Using machine learning to solve your business problems
Machine Learning (ML) is absolutely everywhere. The big three cloud service providers—AWS, Azure, and Google Cloud—have a ton of different machine learning services, with more on the way.
You have a feeling that this approach can help you. But where do you start?
I had the exact same question.
Thankfully, I was able to ask my colleague Nithya Manickam. She’s a principal data scientist here a Lacework, a world class expert in machine learning, and was gracious enough to humour my questions in a livestream over on LinkedIn.
What’s your problem?
While the technology itself is exciting, that’s not enough of a reason to make the investment to use it in your business. You need to make sure that you’re going to get a return on your investment.
Success stories around machine learning hint at unique and novel solutions to really challenging problems. Odds are, your business has a few of these problems.
Solve these and you could unlock new levels of success for your business.
But how do you take that first step? Nithya’s advice is simple, “Get your questions right.”
You don’t start with, “I want to use machine learning” but something along the lines of, “I need to figure out what indicators show the customers most likely to struggle in their first three months.”
Without a good set of questions, your machine learning project won’t succeed.
What ML is good at
The next question is about your questions. Are they a good match for machine learning? While it’s an effective problem solving technique, it does have its strengths and weaknesses.
I asked Nithya about this. She’s spent years applying machine learning techniques to some really intractable problems. Her short list; recommendation engines, route finding, natural language processing, or problems with similar structures.
So, why does machine learning work especially well for these areas?
Each of these problems attempt to draw unique connections between a vast array of disparate data points.
Humans can solve these problems too. It just takes a lot more work, isn’t nearly as repeatable, and most likely isn’t worth the investment of time and effort.
What ML is bad at
The success of machine learning models is tied to the data they are trained on. If that data isn’t comprehensive and unbiased, your results aren’t going to be insightful.
In the cases on biased data, they could even be harmful depending on your use case. Nithya provided a clear example, “With women a few years ago, a simple CEO search would come up with all white men, but that is not the portfolio of all the CEOs we have out there.”
That’s an issue that needs to be addressed. Thankfully work has started. Sadly, there’s still a long way to go.
If you’re interested in the challenges around bias in machine learning, research from the Algorithmic Justice League and the resource collection at the AiEthicist.org are great places to start.
The issue of bias in data isn’t limited to significant community issues. If your sales data or maps or language samples aren’t well rounded and well thought out, you’re going to get subpar results.
NIST has started a new effort around algorithmic bias that is also a great reference and worth digging into.
Your first steps
The big three cloud service providers—AWS, Azure, and Google—have a great line up of machine learning services. They follow the same three layer approach to their offerings.
The lowest layer is there for machine learning experts. It consists of low level tools pre-packaged and ready to go. If you know what those are and how to use them, this layer is for you.
Next is for those with a bit more experience. These offers hold your hand through more of the process but still allow you to build and train your own models.
Finally, there are the top layer services. With these services, you provide something—a photo, text, audio, etc.—and the service returns an analysis.
Basically, you ask “What is this?” and the service gives you an answer.
If one of these top layer services suits your use case, you’re all set. A simple command or a visit to the service web page and you’ll be up and running.
As your familiarity with machine learning grows, you will want to build your own customized machine learning solution. To do that, you’re going to need to understand more about the mechanics and mathematics of machine learning.
To get started on that path, there are some excellent, freely available resources online.
Stephanie Yee and Tony Chu have built a wonderful visual introduction to machine learning. This site literally shows you some of the core concepts and how they can be used to answer the questions about your data sets.
With that understanding in hand, Google’s education initiative offers a very approachable set of guides, videos, and other tools to help wrap your head around all aspect of Machine learning.
If you’re looking for a bigger challenge, Andrew Ng has a YouTube course that takes a deep dive into machine learning. Stanford University also has course notes and slides on how to mine massive datasets.
Finally, “Dive into Deep Learning” is an open textbook that plumbs the nuances with a specific area of machine learning; deep learning.
With persistence and a lot of hard work, you can increase your machine learning knowledge and drive real business results.
Be sure to watch the full discussion with Nithya Manickam and be sure to follow Lacework over on LinkedIn for more discussions like this in our series, Life’s A Breach.