In the News
How Tesla's autopilot learns
Tesla has been rolling out auto pilot features to its cars over the last few days. This is not real autonomous driving yet, but Tesla has big plans and this article details how they intend to use existing Teslas to bolster their efforts.
MIT system that replaces human intuition with algorithms outperforms human teams
Choosing "features" to analyze data usually requires some human intuition. MIT researchers aim to take the human element out of big-data analysis, with a new system that not only searches for patterns but designs the feature set, too.
The challenges of building AI apps
An increasing number of apps take advantage of some form of AI. What challenges does this cause in terms of software, design and data access?
Also in the news this week...
- Google invests in Chinese AI start-up Mobvoi that works on voice-controlled software
- Tim Cook of Apple says a "massive change" is coming to the auto industry
- Sentient technologies, world's most funded AI company, launches online shopping product
- Japan's deep learning startup Alpaca raises $1M to launch AI-based forex trading platform
- Russian military firm claims it has produced technology to enable its military robots to make decisions without human intervention
- Arsenal buys US-based StatDNA to bolster it's scouting and game preparation with big data
Learning
Data Infrastructure at IFTTT
Data is critical at IFTTT. Their pipeline includes Spark, Elasticsearch, AWS RDS, Pipeline, Kafka and more. And it all has to stay flexible, scalable and available. Interesting challenge!
Stanford Tutorial on Deep Learning - Part 2
This part jumps right into Autoencoders, Convolutional Neural Networks and Recurrent Neural Networks. Part 1 on non-linear classifiers and the backpropagation algorithm is available here.
Large scale distributed Deep Learning on Hadoop clusters
Yahoo's work to progressively leverage their Hadoop clusters to train and run complex neural networks.
A quick tour of Torch internals
Torch is a popular open source library to run machine-learning applications. If you've ever wanted to get a better understanding of what Torch does and how it works, have a look at this post.
Software tools & code
Word RNN
Implementation of a Recurrent Neural Network that predicts word-by-word. The author made a fun demonstration of the concept with a clickbait title generator (previously covered) and a fun clickotron.
The library is a useful fork of Karpathy's character-by-character prediction RNN.
Im2Calories demo
Video demo of Google's project to count calories with photos of the food you eat. The interesting insight here is that they don't simply rely on the photo, they take also into account contextual information such as your GPS position and direct input.
Recurrent latent variable model for sequential data
This is an implementation of the paper "A Recurrent Latent Variable Model for Sequential Data", written mostly in pure Theano.
Brains & Neurons
How the brain controls sleep
Sleep is usually considered an all-or-nothing state: The brain is either entirely awake or entirely asleep. However, MIT neuroscientists have discovered a brain circuit that can trigger small regions of the brain to fall asleep or become less alert, while the rest of the brain remains awake.
Some thoughts
The search for the God algorithm
So far no one has discovered the Master Algorithm that can reap and sort all knowledge, but not for want of trying. A new book details the hunt for the modern philosopher’s stone.
Breaking the barrier of humans and machines
To become broadly used and accepted, innovative & automated services will need to understand us better and become more intimate.
How to become a unicorn Data Scientist
What makes a good data scientist? And if you are a good data scientist, how much should you expect to get paid?
About
This newsletter is a weekly collection of AI news and resources. If you find it worthwhile, please forward to your friends and colleagues, or share on your favorite network!
Share on Twitter · Share on Linkedin · Share on Google+
Suggestions or comments are more than welcome, just reply to this email. Thanks!