The Real Status of Today’s Machine Learning

In our last installment, we looked at how small and mid-sized enterprises and machine learning newbies could get started in using this exciting technology. It’s not as hard as it looks, and machine learning services are making it very accessible. Need another bit of encouragement? You won’t be far behind most in the field.

We may all be talking about neural networks, but “[t]he state of the practice is less futuristic,” opines TechCrunch. Most applications of machine learning, even among the tech leaders, are using the same algorithms and engineering tools from years ago. Regression analysis, decision trees, and similar methods are driving ad targeting, product recommendations, and search results ranking to a greater degree than sexy “deep learning” advancements.

What’s more, there are infrastructure issues yet to be solved. The majority of time devoted to machine learning is spent preparing and monitoring the learning tools. Building the AI is a relatively small part of the picture.

Unfortunately, preparing data is a hassle, and the “bigger” the data, the worse the problems. Using scripts to consolidate duplicates, normalize metrics, and so on, can involve days of manual labor for a single run.

Big data can also lead to big machine learning errors, so monitoring production models is essential. Again we reach an impasse: When moving into unsupervised machine learning, where the correct output isn’t known in advance, traditional testing and validation tools no longer work. So how is IT to determine if the model is making “good” predictions? Dashboards and program alerts fill the gap at the development level, and more capable and specific tools are finally being developed by a few innovators.

The point is that machine learning isn’t breaking any molds for rapid adoption. To the contrary, it’s experienced a slow rise. Neural networks joined the scientific literature in the 1930s, the math was completed by the 1990s, and it’s taken the intervening decades for computers to catch up.

The next obstacle will be developing end-to-end solutions, which will accelerate the transition from the rudimentary machine learning dominating business today to the more futuristic possibilities still mostly dormant in neural networking laboratories. How long such a transition will take is still up for debate.


Add Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.

EMT Lead
How to Become an EMT and the Salary to Expect
Martin Richenhagen AGCO CEO
10 Things You Didn’t Know about AGCO CEO Martin Richenhagen
Terry Crews Brooklyn Nine Nine
How Terry Crews Achieved a Net Worth of $20 Million
University of Washington
20 Most Notable University of Washington Alumni in Business
Credit cards in wallet in back pocket
7 Elements of a Good Credit Application
budgeting in the office
Three Ways Budgeting Our Money Actually Makes Us Richer
thermometer
10 Pharmaceutical Stocks to Consider in 2019
DJIA
The History and Evolution of the DJIA
Uptime
Prioritizing Security to Future-Proof Enterprise Collaboration
Does Artificial Intelligence Have Ethics?
Cloud Attack
Top 8 Cyber Security Trends in 2019 You Should Look out For
ClimaCelll
Why ClimaCell Could be a Game Changer in Weather Forecasting
Downtown Dallas
20 Awesome Free Things to Do in Dallas
Four Mexican Getaways You Need to Try At Least Once
Family Friendly Vacation to Dillion, Colorado
Denver
20 Awesome Free Things to Do in Denver
The 20 Most Fuel Efficient Cars in 2019
2013 Nissan GTR
The 10 Most Expensive Cars to Maintain in 2019
The 20 Best Chevy SUVs of All Time
Ferrari 458 Speciale
8 Cars That Have Big Investment Potential in 2019
Timex MK1 Steel Watch With White Dial
The 10 Best Timex Watches of 2019
Fossil Sport
The 10 Best Fossil Watches of 2019
Hamilton Watches feature
The 10 Best Hamilton Watches of 2019
The 20 Best Tourbillon Watches in 2019