“Nvidia’s cutting-edge graphics cards used for playing video games and–increasingly–for performing artificial intelligence and machine learning tasks can cost hundreds or even thousands of dollars. But on Monday, CEO Jensen Huang announced a product for the “maker” market of students, hobbyists, and tinkerers with a tiny $99 computer optimized for A.I and machine learning.
Dubbed the Jetson Nano, the 3-inch by 4-inch computer includes a graphics processor from Nvidia and a standard processor based on designs from ARM. Requiring little electricity to run so it could be used with…”
Source:
http://fortune.com/2019/03/18/nvidia-jetson-nano-maker-diy/
I will certainly buy a few. As a general PC, for home entertainment and AI coding/research. Pity it doesn’t have wi-fi built in. That’s the only negative.
Someone could make a sub $200 laptop out of the module and I certainly would be in the market for that too.
Slightly off topic:
Can you have a neuron with less than one weight? I think you can with what Google has decided to call spinners:
https://github.com/FALCONN-LIB/FFHT/issues/26
Random projections and more structured projections like spinners allow for multiple possible weight sharing options with neural networks. You can also organize things more as deep extreme learning machines that as deep neural networks allowing even more opportunities.
https://www.facebook.com/yann.lecun/posts/10152872571572143
In fact extreme learning machines are a simple form of associative memory.
However if you use multiple layers of them they are as expressive as deep neural networks, or may indeed have some advantages over them.