There are a couple of seemingly contradictory memes rolling around the deep learning field. One is that you need a truly epic amount of data to do interesting work. The other is that in many subject areas there is a ton of data but it's not just laying around for data scientists to snarf up.

On this week's Structure Show podcast, Enlitic Founder and CEO Jeremy Howard and Senior Data Scientist Ahna Girshick address those topics and more.

Girshick, who is our first guest who's worked with Philip Glass and Björk on creating music visualizations, said while there may be scads of MRIs, CAT scans, x-rays created -- they're typically used for their primary purpose -- to diagnose your bum knee, and are then squirreled away in some PACS system never to see the light of day again.

All of that data is useful for machine learning algorithms, or would be, if it were accessible, she said.

Girshick and  Howard agreed that while deep learning -- the process of a computer teaching itself how to solve a problem -- gets better with more data, there's no reason to hold off working with it for data become available.

"While more data can be better I think this is stopping people from trying to use big data inappropriately," Howard said. He cited a recent Kaggle competition on facial key point recognition that's  using 7,000 images and "the top algorithms are nearly perfectly accurate."

The reason companies like Baidu and Google say you need mountains of data is because they have mountains of data available, he said.  "I don't think people should be put off trying to use deep learing just because they don't have a lot of data."

Enlitic is using deep learning to provide medical diagnoses faster and help provide better medical and outcomes for millions of underserved people.

It's a fascinating discussion so please check it out -- Girshick will speak more on what Enlitic is doing at Structure Data next month.

And, if you want to hear what's going on with Pivotal's big data portfolio, Derrick Harris has the latest. Oh and Microsoft makes a bold play for startups by ponying up $500K in Azure cloud credits starting with the Y Combinator Winter 2015 class. That ups the ante pretty significantly compared to what Amazon Web Services, Google and IBM offer. Your move boys.

No, you don't need a ton of data to do deep learning

The Cure for Information Overload
http://www.attentionfusion.com

I'm machine learning enthusiast with applied math background and a lot experience as full stack developer.
I'm from Ukraine but now I'm working remotely in Montenegro.

A weekly podcast from GigaOM providing the latest news, analysis and interviews about the cloud, infrastructure and big data spaces. 

No, you don't need a ton of data to do deep learning

Gigaom Structure