The machine learning sub-discipline of study in Artificial Intelligence has benefited from burgeoning attention in the past few decades. Machine learning has been applied in recent years to a gamut of conceivable problem spaces. Some applications have pervaded cultural awareness in obvious ways, such as self-driving cars, facial recognition, targeted advertising, and product recommendations. For […]
The machine learning sub-discipline of study in Artificial Intelligence has benefited from burgeoning attention in the past few decades. Machine learning has been applied in recent years to a gamut of conceivable problem spaces. Some applications have pervaded cultural awareness in obvious ways, such as self-driving cars, facial recognition, targeted advertising, and product recommendations.
For those of us working in the technology industry, there are now tools and precedents for using machine learning to solve highly complex and important industry challenges like understanding customer sentiments, classifying data to better understand and use it, and personalizing search experiences.
When attempting to solve problems with machine learning it is important to have a well-defined problem, a careful approach, and a good justification. Here are four reasons why:
1. Data and resource consumption are significant
Most types of machine learning techniques rely on large training data sets. Finding the data to accommodate all of this is sometimes not feasible, and when it is, a lot of resources must be obtained for storage and processing.
2. The provided tools aren’t always able to encapsulate their complexity, and it is easy to get in over your head
When you use a machine learning algorithm, you are mostly dealing in the input and output of someone else’s very complex function. And that may be actively changing even as you use it. It is possible to ignore this fact and rush to get something to work without fully understanding it, but this proves perilous if you run into problems.
3. Machine learning presents a multidisciplinary challenge
Machine learning depends on a coordination of specialized knowledge areas. To make good decisions about your approach, you have to answer questions such as “how much processing power do I need?”, “Is this the right algorithm to solve my problem?”, “what parameters should I use?”, and “has my model converged yet?” These questions have interdependencies on each other, and to answer all of them requires a varied background.
4. Machine learning is not always the right solution
In many cases, machine learning is applied to problems because they take too long or are too complex to solve without it. Using it in cases where an alternative exists may prove to be a mistake. Machine learning deals in uncertainties, and though these uncertainties are whittled down by an enormous amount of input over several iterations, they are still uncertainties. If you have a way to know what is exactly right for you, make sure the way of getting to it is actually slower or more costly before choosing a solution that relies on chance.
Even with these caveats, machine learning has already been used to better our lives in more ways than I can list and the best way to avoid pitfalls with it is to get started learning and trying it out!
Considering machine learning? Tell us what you think!
The National Institute of Standards and Technology (NIST) recently proposed four principles for explainable artificial...
Last fall, Excella participated in the Department of Defense’s (DoD) Eye in the Sky Challenge....