Saturday, April 13, 2019

How To Get Started In AI For Less Than $400


Got 400 bucks and some free time? Then you may be the next artificial intelligence (AI) whiz kid.

It’s never been easier for a developer to start building something in the field of artificial intelligence. Firms are creating total hardware solutions that run in the hundreds, not the hundreds of thousands, of dollars. Much of the software and frameworks are free and open source. Data sets to train your algorithm? Also free and publicly available. Unlimited computing power and prebuilt models you can tap at the touch of a button? Not free, but cheap compared to building your own data center. Thanks, public cloud.

But what does all of this mean for the future of technology? As we found out from speaking to AI developers at Intel’s AI DevCon in May 2018 in San Francisco, a great deal. The last startup boom came about as a rising supply of talented software developers met an influx of cheap computing power from the cloud that could be turned on and off like water from a tap. This gave developers the ability to go out and build something if they had an idea, versus raising millions to get a proof of concept going. The shadows of those early hackathons loom large over today’s economy as many mega startups and tech titans were born as a result.

If AI is tech’s next frontier—as almost everyone not living under a rock seems to agree—then the ability for developers to go from idea to proof of concept for mostly the cost of their own time could be an even bigger game changer. In this article, we’ll look at some of the technologies that are making it easier than ever for developers to get started.

Key Innovations

Computer chips’ performance is continuing to rise, and their prices are still falling—all at an exponential rate. The falling price and rising impact of hardware is a key catalyst of this new wave of AI experimentation. Let’s take a look at some of the specific innovations taking place in this space.

Image Recognition And Machine Vision Innovations: Amazon launched a nearly complete AI visual development device in June. Called AWS DeepLens, it gives developers all the tools they need to build an image recognition/ machine vision application by combining a camera, system board and integral microphones into a small package. At just a couple hundred bucks, the price point makes it easy for even novices to give it a go—especially since AWS DeepLens comes with demo projects and apps that are ready to work 10 minutes after taking it out of the box. No big corporate expense account required to play here.

Putting AI At The Edge With Ultra-Low Power: Many of the most interesting uses of AI involve those workloads being processed on specific devices at the edge of a network. An example of this would be a camera attached to a drone that is trained to detect obstacles and to avoid them in real time, without sending the data through the internet to a data center for processing. But such devices require high performance at low power; otherwise, you’d have to recharge the batteries every few minutes. 

Intel has a solution that looks like a flash drive and comes with an extensive library of demo code and support tools. Like a flash drive, it’s even designed to plug into a USB port on a laptop, Raspberry Pi or other device. Called the Movidius Neural Compute Stick, it’s useful for developers who are leaning about AI or for more experienced pros looking for a research platform. This essentially puts AI-grade silicon in the hands of anyone who wants it.
uncaptioned
The Raspberry Pi 2 is the second generation, replacing the original. jessekarjalainen/iStock
The OG Of Programmable Hardware: A ubiquitous presence in any developer or tinker’s arsenal since the first version was created in 2012, the Raspberry Pi is a small, credit-card sized, low-power computer on a single board (or system on a chip—SOC). Designed by a nonprofit foundation interested in promoting basic computer science skills, the device has been the basis for countless new device designs and use cases thanks to its low cost ($40), extreme ubiquity and ease of use. 

A major hotel chain, attempting to better assess guests’ experiences in real time, attached a microphone to a Raspberry Pi and hooked it up to the front desk. Machine learning software analyzed the tone of guests’ voices (but not the words, for privacy) to see if they were pleased or displeased. While it may need other components to function as a full solution, this can be a backbone for many new device concepts.

Free Hardware: Developers often say the best hardware is the hardware you don’t have to pay for. How do you get free hardware? That’s where the cloud comes in. Cloud providers install, maintain, upgrade and manage the hardware and lease all the computing power—and now, pretrained AI models—you need for a simple low price. Because AI is one of the ways cloud providers are seeking to differentiate themselves and provide more value to customers, the offerings in terms of infrastructure, applications, data libraries and models are only set to improve over time.

Open Source: The Way AI Is Done

Through the hard work of many talented AI professionals, it’s increasingly possible to create applications using only the common language Python versus something more specialized. As an open-source language, Python is the most common skill among developers (it’s even used to teach children), so this effectively opens up the ability to work on AI to anyone with programming experience. 

In addition, many critical frameworks for developing deep learning, such as Caffe and TensorFlow, are open source. That’s not all. Many deep learning models themselves are proudly open source, so you can deploy them as is or use them as a building block to build a more custom solution.
The falling price and rising impact of hardware is a key catalyst of this new wave of AI experimentation.
But if AI is so valuable, why are firms just giving away this intellectual property? 
First of all, AI is so new and so complex that it really does take academics and businesspeople putting their cards on the table and comparing notes to get the advancements we need. It also helps when all developers are working from the same playbook so the wheel isn’t being reinvented. 

Moreover, many of the deep learning models themselves could theoretically be reverse engineered anyway. The data used to train these models, though? That can’t be reverse engineered, so many firms choose to keep that proprietary.

But not all organizations do. Many cloud providers, nonprofits and governments are providing labeled input/output data sets that developers can use to train models. In fact, one of the pioneers of AI who now works at a social media giant says most research and development really takes place on these publicly available data sets, not the proprietary corporate ones that would seem a big barrier to entry. 

Spending Your Time Wisely

There’s a significant time investment involved in learning to use the tools and libraries that have been created to implement AI projects. This is perhaps the biggest remaining roadblock to any enthusiast who wants to dive into this space. Intel and other AI hardware manufacturers are actively developing APIs (application programming interfaces), toolkits and code libraries that simplify the tasks involved in programming AI. These tools can save the developer hundreds (perhaps thousands) of hours developing code that already exists. There are also robust user groups and developer sites that provide support for software and hardware development.
If AI is tech’s next frontier, then the ability for developers to go from idea to proof of concept for mostly the cost of their own time could be a game changer.
Still, if you want to be an AI developer, you just need to take the time to learn the trade. A variety of resources are available online for free. YouTube features myriad how-to videos on the topic. Or, for a more academic approach, Coursera, edX, Udacity and General Assembly offer courses on the topic ranging in price from free to a few thousand dollars. Certainly, though, if you feel motivated to get your master’s or a Ph.D. in the space, we won’t stop you. Ph.D. data scientists with a foundation in machine learning are among the most highly paid professionals in the world today.
The moral of the story: If AI excites you and you think it’s something you’d like to do, or at least explore, then just get started. It’s never been cheaper or easier to do so.
CREDIT: Pekic/iStock



About the Author
You may know us for our processors. But we do so much more. Intel invents at the boundaries of technology to make amazing experiences possible for business and society, and for every person on Earth.
Harnessing the capability of the cloud, the ubiquity of the Internet of Things, the latest advances in memory and programmable solutions, and the promise of always-on 5G connectivity and artificial intelligence, Intel is disrupting industries and solving global challenges. Leading on policy, diversity, inclusion, education and sustainability, we create value for our stockholders, customers, and society.


No comments:

Post a Comment