Deep Learning Systems, Artificial Intelligence and Cloud Computing

Are we experiencing a revolution in deep learning and artificial intelligence based on the power of cloud computing?

Advances in deep learning and artificial intelligence are accelerating because massive computing power is finally accessible to companies of all sizes. Cloud computing is proving itself a true game-changer in this “futuristic” sector because it affords a variety of critical resources to, arguably, the most creative people in the world—entrepreneurs.

One of these entrepreneurs who thinks differently is Jason Toy. He is the founder and CEO of Somatic, a platform for anyone to easily build deep learning applications. He has been building software companies and products from the ground up for 10 years. Jason considers himself a generalist who works wherever the core problems are: programming, sales, training, management, marketing, and DevOps.

I recently put some questions to him about the topics that are most dear to him: deep learning and artificial intelligence.

Before we begin the interview proper, Jason, would you explain how you are using the terms deep learning, machine learning, and artificial intelligence? I sometimes see these terms used interchangeably and that seems imprecise.

Artificial intelligence typically refers to computers that learn to do things instead of being told explicitly what to do. There are many different types of AI systems, knowledge systems, statistics-based systems, and ontology systems for starters. Any computer system that looks like it is smart can be considered AI.
Machine Learning is basically a subset of AI and means that the algorithms and models are statistically based. Machine learning is synonymous with statistical modeling.

Deep Learning is a subset of machine learning, and its just a fancy new name for neural networks. Neural Networks have been around for decades, but due to recent advances in hardware, we can create “deeper” neural networks, or add more layers.

You started a number of companies. Why are you working on Deep Learning now?

Artificial Intelligence has been one of my main interests since I started programming in high school. In college, I studied math and computer science. I started a company (SocMetrics) a few years ago using machine learning to help organizations better understand their customers.
What is interesting to me about deep learning is that the algorithms are able to automatically extract features from the data.

With more classical machine learning methods, you as an engineer or data scientist, had to manually code up the features that are fed into the machine learning algorithms. That results in a lot of resources consumed into something that might not even be done correctly.

I got a renewed interest in machine learning when deep learning became big, because I saw the pathway to much more intelligent machines. Deep Learning is an incredibly hot subject now, and some of it is hype, but there are amazing things the technology can do that was never possible before, like self-driving cars and accurate image recognition.  But how can you get more companies to adopt these technologies? Right now to do deep learning you need a team of people who are trained in data science,  developers, and neural networks.  Education will immensely help, but there also needs to be tools that anyone can use. With Somatic, I hope to make the building of deep learning systems available to everyone interested, including scientists, business people, students, artists, and amateurs. Imagine what if you had an idea for a deep learning model, and you could get an initial prototype up by yourself in hours instead of months.

The cloud is making resources available and scalable for smaller and emerging organizations. How do you see what you are working on relating to cloud computing?

deep learningWell, my ultimate goal is to open up machine learning to everyone. That means several things, such as creating easier tools for machine learning, education around machine learning, and better infrastructure around machine learning.
If companies and amateurs want to utilize deep learning, then the fastest way to do that is using on-demand cloud resources from AWS, Azure, Google, etc. With Somatic, we aim to utilize public cloud infrastructure to help people build machine learning applications faster and more efficiently. And honestly, we would not be able to build what we are trying to build without the public cloud infrastructure companies.

You blog and comment about computing and scientific trends. Do you see a role for online education providers?

Absolutely, I love platforms like Cloudera and Mitx that are opening up education to everyone. There are schools that are opening up courses related to deep learning all over the US. The field is moving so fast, though, and so I think online education providers will have an advantage because they can update their materials faster, try out new strategies, and a/b what is and isn’t working. For example, what is interesting about Cloud Academy to me is the hands-on labs where students can test their skills with real software instead of contrived tests.

I see a future where more education around deep learning is happening online. I would love to see online deep learning courses where robots are trained up with different algorithms and data sets. Students could train their models on these platforms and then set them loose in a virtual environment to see how they are progressing. I believe that the creativity humans have will help create new and wonderful intelligent applications. That means we need to get more people educated on how deep learning works and how they can build these systems, and so it’s an issue I think about often.

Do you have any predictions for how Deep Learning and cloud will be used in the future?

Deep learning requires heavy computing resources. It is cost prohibitive to build the infrastructure yourself and power it locally. Like everything else in computing, deep learning will also be in the cloud to utilize the massive infrastructure available online. And so I see a lot of companies building and investing in deep learning systems on the cloud. On the other side, there will be a lot of deep learning technology embedded into our local devices.

Apple is working on getting their software like Siri to work on their phones when there is no internet connection. NVIDIA has released special versions of their hardware to work in cars and mobile devices to specifically power deep learning applications. In the future, we will literally be surrounded by intelligent computer systems regardless of being in the cloud or local.

Google Vision API: Image Analysis as a Service is a popular post related to much of what Jason Toy discussed above. That post, and another one on Amazon Mechanical Turk were written by Cloud Academy Sr. Software Engineer, Alex Casalboni.

Cloud Academy has an advanced course, Introduction to the Principles and Practice of Amazon Machine Learning, written by James Counts that may interest you. It is just over 2 hours and anyone can sign up for a free 7-day trial to evaluate this or any other course, lab, or quiz we offer. Thanks for reading and as the editor here I’d like to encourage you to seen feedback so we can continue growing and learning.

Cloud Academy