The NVIDIA Jetson Nano

During the last few weeks, I’ve had the opportunity to play around with the NVIDIA Jetson Nano. As an artificial intelligence enthusiast, I was very eager to see what the Jetson Nano could do, and I was not disappointed.

The Jetson Nano is a small AI computer, about the size of a Raspberry Pi. Just like the Pi, the Nano runs Linux and contains a 40-pin header for interfacing with other components. With 4 USB ports, ethernet, and HDMI, it’s a self-contained computer. What separates the Nano from the Pi, however, is the NVIDIA GPU. With 128 cores, the GPU is capable of performing floating point operations substantially faster than a typical machine. It’s those very floating point calculations that are the bulk of AI processing applications.

Taking the Nano for a spin, I downloaded their OS image and installed on an SD card. Visiting NVIDIA’s website, I found that they offered a free online course to learn the basics. Through their course, I was able to create simple AI to perform real-time image classification. Unlike a typical Haar filter generation, the Nano was able to create a classifier in minutes. And, even more exciting, the PyTorch framework they demonstrate contains pre-trained networks that allow for far fewer images than a typical Haar filter requires. Additional NVIDIA courses include video processing using C++, medical image processing, and countless other AI tasks.

What makes the Nano so exciting is just how fast it works. Having developed AI models in the past, I’m accustomed to lots of waiting. However, the Jetson Nano functions like a mini AI supercomputer. For beginners, this means you can easily generate your neural network model quickly and perform validation tests with substantially less effort than traditional computer architectures.

I highly recommend the Jetson Nano for anyone wanting to learn more about artificial intelligence or wishing to experiment with potential consumer or industrial applications with computer vision.

AI Resources

Today, there are countless AI products and resources available for the developer. I’d like to review a few of the resources I’ve used.


The first resource on my list is AI For Humans (AIFH) by Jeff Heaton. Of all the books on programming artificial intelligence, his books are by far the best. Written without any heavy math, Jeff’s books explain just about everything you could want to know about artificial intelligence. Additionally, Jeff has his own framework called ENCOG which can be used in Java, or can be run as a standalone GUI for development. Of all the resources I’ve used to date, this is without hesitation the best.

AWS DeepLens

I was recently sent an AWS DeepLens by a client for an artificial intelligence project. It’s always exciting to receive hardware from a client, so I was certainly looking forward to this! Unfortunately, my experience with it has been a bit less exciting. For starters, I’m stuck trying to figure out what exactly the purpose of this box is. Am I to use this to learn artificial intelligence? Is this box intended to be embedded within a product? I have absolutely no idea. While the DeepLens is a neat toy, the setup is far from simple. An internet connection to AWS is needed, certificates need installed on your machine, and everything is configured through your Amazon account. Even worse, the service does not appear to be free. (Note, you do get one free year on AWS, but after that you pay for the service. After that, I have no idea what it costs to run the DeepLens, but I’ve noticed warnings about costs associated with some actions). I applaud Amazon for trying to bring deep learning to the masses, but I think this product is a dud.


DeepLearning4Java (DL4J) is a Java library for deep learning AI. Written by SkyMind, DL4J is one of the most well-known AI libraries for Java. With the earlier versions of DL4J, the user would have to install various native libraries such as ND4J (N-Dimensional Arrays 4 Java). This proved to be more difficult than it sounds as various libraries were dependent on other libraries, documentation was scarce, and error messages were cryptic at best. Fortunately, with the 1.X version of DL4J, the install process is streamlined to sampling cloning a repository and running a maven build. Native libraries are managed within the maven build, saving users the trouble from earlier versions. With these changes, DL4J is an excellent framework I would recommend for any AI project.

Getting Started with Artificial Intelligence

It seems that artificial intelligence is in the news more and more. Most larger companies use AI for something within their business, and more and more businesses are finding ways to improve their organization with AI. Purchase recommendation systems, self driving cars, video games, language translation apps, and route mapping software are just a few examples of artificial intelligence we see every day.

But where does someone interested in AI get started?  AI can be very complicated, but that doesn’t mean you can’t get involved too! One of my favorite resources for learning AI is the series Artificial Intelligence for Humans by Jeff Heaton. In his books, he covers a variety of topics including genetic algorithms, machine learning, clustering, linear regression, swarm algorithms, and so much more. While these topics can be complicated, Jeff presents them without all the math in a way that is far more readable than most texts on the topic.

If you are interested in AI, one of the first questions to answer is what programming language you want to use. While any language can be used for AI, the bulk of the tools and frameworks available exist for Java, C++, Python, and R. If you’re big into number crunching and Big Data, R may be the obvious choice. If you’re not a programmer, Python may be easier. For existing developers, Java or C++ may be best.

What type of AI do you start with? Try genetic algorithms or swarming algorithms. Genetic AI assumes the answer to be like a genetic genome and, through a series of mutations and genetic splicing, attempts to find an answer.  Swarming algorithms look at groups of objects and attempts to have them behave like a cohesive team. Swarming algorithms are great for games with AI controlled enemy armies and are commonly used by game developers. Other common, and simple, AI algorithms include K-Means clustering (used to group objects by similarity),  linear regression (used to predict unknown values using relatively simple algebra), or path finding algorithms such as Dijkstra’s Algorithm.

Once you understand the basics of AI, you can move onto frameworks like DeepLearning4J or TensorFlow to help create Neural Networks (a far more advanced type of AI) or look into libraries like OpenCV for tinkering with Computer Vision.

Whether it’s creating an emulation of a fish tank using swarming algorithms, solving the Traveling Salesman Problem using genetic algorithms, or calculating a path through a maze using Dijkstra’s Algorithm, artificial intelligence is loads of fun.