Machine Learning Isn’t Magic

From the Voices on Project Management Blog
by , , , , , , , , , , , , , , , , , , , , , , ,
Voices on Project Management offers insights, tips, advice and personal stories from project managers in different regions and industries. The goal is to get you thinking, and spark a discussion. So, if you read something that you agree with--or even disagree with--leave a comment.

About this Blog

RSS

View Posts By:

Cameron McGaughy
Marian Haus
Lynda Bourne
Lung-Hung Chou
Bernadine Douglas
Kevin Korterud
Conrado Morlan
Peter Tarhanidis
Mario Trentim
Jen Skrabak
David Wakeman
Roberto Toledo
Vivek Prakash
Cyndee Miller
Shobhna Raghupathy
Wanda Curlee
Rex Holmlin
Christian Bisson
Taralyn Frasqueri-Molina
Jess Tayel
Ramiro Rodrigues
Linda Agyapong
Joanna Newman
Soma Bhattacharya

Past Contributers:

Jorge Valdés Garciatorres
Hajar Hamid
Dan Goldfischer
Saira Karim
Jim De Piante
sanjay saini
Judy Umlas
Abdiel Ledesma
Michael Hatfield
Deanna Landers
Alfonso Bucero
Kelley Hunsberger
William Krebs
Peter Taylor
Rebecca Braglio
Geoff Mattie
Dmitri Ivanenko PMP ITIL

Recent Posts

3 Project Management Lessons From a 70.3 Ironman

Follow These 3 Steps to Validate a Variance

Unlock the Value of Artificial Intelligence

Stakeholder Management for Traveling Families

3 Tips For Cultivating Your Executive Presence


Categories: Innovation, IT, Lessons Learned, ROI


By Christian Bisson, PMP

Machine learning is one of today’s hottest tech topics.

It’s essentially a type of artificial intelligence (AI) in which you give your software the ability to “learn” based on data. For example, you probably notice how YouTube, Netflix, Amazon and many other companies suggests videos or products you should check out. These suggestions are based on your previous online actions, or those of other people deemed “similar” to you.

For some time now I’ve been working on projects that involve this technology. We often have clients who want machine learning even though they do not know if it’s even relevant to them. Since “everyone is doing it,” they want to do it too.

Calibrating a project sponsor’s expectations is often a good idea. While the automated services generated through machine learning may seem magical, getting to that point involves challenges—and a lot of work.

1. It needs quality data.

The machine will learn using the data it has being given—that data is the crucial starting point. The data that’s available is what drives how the machine will evolve and what added value machine learning can bring to your project/product. For example, if you are trying to teach the machine to recognize vehicles on images it scans, and all you can teach it with are images of small cars, you are not set up for success. You need a better variety of images.

The machine’s ability to learn is directly tied to the quality of the data it encounters.

2. It needs lots of data.

Once you have quality data, you need it in high quantities. If you can only provide the machine with the website behaviors of, say, hundreds of users per month, don’t expect it to have enough information to be able to recommend the best products based on user trends. Its sample will be too little to be able to be accurate.

3. It needs to be tested continually.

Once you have the necessary data, the journey is not over. The machine may learn on its own, but it’s learning based on how it was built and with the data it’s being fed. There is always room for improvement.

4. It’s costly.

As amazing as machine learning is, it is not cheap. So keep an eye on your project’s budget. Machine learning experts can command high salaries, and there is a lot of effort involved with researching the best approach—creating the models, training them, testing them, etc. Make sure the ROI is worth it.

Have you had a chance to work on a project involving machine learning? What challenges have you faced?

Posted by Christian Bisson on: July 14, 2018 08:59 AM | Permalink

Comments (12)

Please login or join to subscribe to this item
Quality in, quality out. Data information knowledge is such a finely balanced process to get value from it. Thanks Christian.

Thanks, Christian. Absolutely.

Quite a concise introduction! thanks for sharing.

I don't think it necessarily needs to be costly. You can get some good feedback about your dataset and possibly find some good data science candidates by posting your project to a community like kaggle.com. The downside is that you have to be open to sharing your data with the community which may contain sensitive info.

Good prelude on machine learning. Thanks for sharing.

Thank you. The content gives a better insight .

Thanks everyone for your feedback.

Al chen, I agree that their might be ways to reduce the costs, but by experience, reducing costs generally comes with less quality or other constraints (ie. sharing data like you mentioned) so it is something to keep in mind if it's a deal breaker or not for the project!

Cheers.

There was a good presentation on AI / Machine Learning at the Dubai International Project Management Forum yesterday, and it is true that if you don't rely on the cloud to mutualize LOADS of data, what you can achieve with it is quite limited...

Please Login/Register to leave a comment.

ADVERTISEMENTS

"Truth comes out of error more readily than out of confusion."

- Francis Bacon

ADVERTISEMENT

Sponsors

Vendor Events

See all Vendor Events