This Week in AI, January 25, 2018
Machine Learning in the cloud from Google, Neuroevolution with Uber, Coachella’s lineup by way of neural network, and more!
Google Launches AutoML
Google recently launched AutoML, a cloud service simplifying machine learning models such that any business can take advantage of recent advancements.
This is a great step forward in bringing AI models to every business and organization. But I think there will be an increasing demand for machine learning models that run natively on our hardware rather than being shipped off to the cloud. This will be great for internal data analysis and tooling, but less important for new products built around AI.
Deep Neuroevolution Sees Success
Uber has released a suite of papers detailing the use of neuroevolution for training deep neural networks on reinforcement learning tasks. Neuroevolution uses genetic algorithms to train networks, seeing better results than common models like deep Q-learning and A3C.
Microsoft Builds an AI that draws for you
The AI team at Microsoft combined attention and GANs (a model appropriately called AttnGAN) to generate images from a string of words. There has been a lot of effort getting recurrent networks working with GANs, so it’s nice to see some progress here. Check out the paper too!
PyTorch models visualized in TensorBoard
One of my favorite things about TensorFlow is being able to visualize your models and parameters in TensorBoard. PyTorch has been missing this functionality, but no longer! Now there’s a TensorBoard extension for viewing your PyTorch models.
You can find a live demo here and the code itself on GitHub.
A Year in PyTorch
January 19th was the one year anniversary of PyTorch going public. In that time it’s grown to be one of the most popular deep learning frameworks. The team has made a lot of progress in a very short amount of time, really pushing the performance and utility of the framework. I’m looking forward to what they accomplish in 2018.
As Seen on Twitter
I’ve realized recently that text generated with RNNs hits the sweet spot of the uncanny valley. The text is similar enough to actual language to seem real, but it’s off just enough to be odd and funny. What’s your favorite act?
~
Stay tuned for new updates as we continue to review all that’s new in the world of AI! And if you’re interested in mastering these transformational skills, and building a rewarding career in this amazing space, consider one of our Nanodegree programs:
