AI Tools Are Popping Like Popcorn

Daniel Shapiro, PhD
3 min readSep 17, 2017

--

I want to take a break from talking business to talk a bit about the technology story that’s unfolding in deep learning.

There are quite a few tools out there, not to mention all the APIs. I have a preferred set of stuff to work with, but it’s fun to see all the stuff coming out, like popping popcorn, there is just so much going on. I’ve been thinking about how the public sees deep learning, artificial intelligence, and deep learning. I read about this stuff, and talk with engineer and business people who are just fascinated by the topic. There is quite a bit of interest from the general public, and from developers. And why not? It’s a cool topic.

Without warning, DigitalOcean popped out a one-click install with a machine learning image. NICE!

Even without GPUs, you gotta love DigitalOcean.

Things are changing. I’m seeing a quiet tools war coming on between Keras+TensorFlow (Google’s stuff) and Caffe/Torch/Cognitive Toolkit, which are supported by facebook, and microsoft. These tools are all free and open source, and each set of tools has more backers than just these companies. In fact, the user groups are the real drivers of progress now, but I have this feeling that we are in the VHS/Betamax days. A real format war. Ask me 1 year ago, and I would say that the fight was between Tensorflow and Theano, but my, how quickly things change.

And then there is the hardware side of things. AMD has faded into the shadows while nVidia is running the tables in the hardware game. And then out comes Google with the TPU 1, and then TPU 2. This recent analysis from nVidia and MIT of the AI hardware space is deep and fascinating. My background in hardware architecture really helped me soak in all the new tricks they are doing on the machine learning compiler and hardware side. The most shocking part of the talk is on slide 249: Only 10% of the power in a GPU is dedicated to the multipliers (computing), and the rest of the power is wasted shuffling data in and out of those same multipliers to/from memory units. It’s all moving at breakneck pace. But, there will be a winner. We, the makers of stuff, need standards.

This “let’s beat up on Google” thing has been going on for a long time, but it was recently announces that Microsoft and Facebook are in bed together. On the bright side, they will release the code. So at least the playing field for us little guys is not going to be a calamity.

It’s a great time to be alive. I can’t wait to see what happens next!

Happy Coding!

-Daniel
daniel@lemay.ai ← Say hi.
Lemay.ai
1(855)LEMAY-AI

Other articles you may enjoy:

--

--