Deep Learning: Intelligence from Big Data

Deep Learning: Intelligence from Big Data

“Any product that excites you over the next five years probably uses this…”
-Steve Jurvetson, Partner, DFJ Ventures

Imagine on-boarding a new hire that works wonders with data at your start-up. Imagine he can work 24/7 and there’s no problem too big or too insignificant for him. Say you’ve got a data set but you can’t find the signal from the noise. Maybe the data needs to be tagged and sorted before it’s analyzed. He can help with that.

And he doesn’t even want options or equity.

Want to iterate your way to a better rocket booster, battery charger, or even a farming implement? He’s got you covered.

And he won’t eat your VC bought organic strawberries. He’s a deep learning algorithm.

There’s only one catch: this new team member- deep learning – won’t tell you why his solutions work. He won’t divulge his secrets. You understand how he gets there (sort of), and you understand his solutions are better. You just won’t be able to understand why they work. Deep learning, the topic of September’s VLAB event, is a bit of a black box, but you’ll want to use it none the less.

What is deep learning and what does it do?

“We can’t really talk about deep learning without talking about data. There’s really two kinds of data: data that’s recorded from the physical world

[images, sound recordings, etc.], and then there’s data that humans produce [e.g., meta data, tagging via data entry, transcriptions, written words, etc.]. Now we’re getting to a world where we can take measurements of the physical world and turn that into symbols that we can search and sort.”
-Adam Berenzweig, Co-founder and CTO, Clarifai

Deep learning algorithms can analyze just about anything – including images. They spot patterns and draw conclusions. Algorithms can transcribe speech and add accurate metadata to images. For consumers, this means no more tagging pictures or sorting email. For businesses, this means making sense of messy data. The algorithm might need some initial guidance to know if it’s on the right track, but it’ll learn to be self sufficient.

Steve Jurvetson describes the deep learning process as a non-linear iteration. The process proliferates and grows today because of a number of factors: vast quantities of (big) data, algorithmic advances that incorporate successive layers of learning, and synapse scale (think number of high-powered processors we can link together).

Operators of a deep learning algorithm feed data to the algorithm and then train the algorithm to evaluate the data effectively. The operators penalize evaluative weights that make wrong tags and reinforce weights that make correct tags until the algorithm can produce reliable results. Deep learning processes have recently disrupted robotics, manufacturing, agriculture, and a plethora of other industries, and deep learning processes will continue to do so.

Author’s aside: Still fuzzy after my ambiguous description of how it works? I am, too. I kept it light because if I ventured into the weeds, I’d invariably stick my foot in my mouth.

So you’re saying I should start a deep learning company?

Probably not. Ilya Sutskever of Google Brain says that deep learning is “easy in the same way that rocket science is easy. It’s easy to understand why a rocket goes to space.” It’s the engineering that’s hard. Elliot Turner of AlchemyAPI does, however, encourage entrepreneurs to think of deep learning as a tool in your set of business applications.

Turner says to first find a complex business problem to solve, and then see if deep learning can help. “How do you make money from deep learning? You transform information to action. ‘Do I sell my stock or close my borders?’ There’s a huge potential to make money.”

This author can dream up a number of interesting applications: acquisition decision making, new recipe creation, creating a pleasant melody or new pop song, controlling physical environments and responding to unforeseen events, knowing when to call the police after an alarm is triggered, etc. (NB: I have no idea if today’s deep learning algorithms can do such things, but I’d like to find out.)

And you don’t have to hire a Ph.D in neural networks to use deep learning…

“From the product design side of things, you need to know what can be done. That doesn’t require all that engineering expertise. We’re getting to the point that that engineering is getting democratized.”
-Naveen Rao

The Ph.D’s and brain scientists are making the tools for you. Clarifai applies award winning speed to image recognition and tagging. Skymind will analyze millions of news articles for you to determine if someone’s talking about your company with emotional bias. AlchemyAPI will help you develop apps to understand customer behavior. Nervana systems develops hardware specifically for deep learning algorithms, and Ersatz Labs provides web based access to deep learning modules so you can run your own experiments.

Have you been wondering what to do with that glut of data on your server, wondering if it’s good for anything? Deep learning might be for you.

This event’s panelists included:
Steve Jurvetson, Partner, DFJ Ventures
Adam Berenzweig, Co-founder and CTO, Clarifai
Naveen Rao, Co-founder and CEO, Nervana Systems
Elliot Turner, Founder and CEO, AlchemyAPI
Ilya Sutskever, Research Scientist, Google Brain

Demo Companies:
Clarifai | SkyMind | Ersatz Labs | AlchemyAPI

Be sure to check out our upcoming events…

Written by Eric McClellan, VLAB Marketing Co-Chair.

By | 2017-05-18T17:47:12+00:00 September 28th, 2014|Event Summary, VLAB|Comments Off on Deep Learning: Intelligence from Big Data