Learning Probability and Statistics

The AI Winter is a distant memory.  Machine Learning and AI are finally making tangible progress. (Siri, Alexa, cars, Wall Street). Statistics is the foundation of the current progress in machine learning.

I started learning some probability and statistics in 2017. I watched the entire Stats 131a class by Michael C Cranston’s at UC Irvine.

I took some high-level notes on Github so I could reference certain topics for a deeper dive.  In addition, I have more complete notes for the first four classes:

Lecture 1Lecture 2Lecture 3Lecture 4

Another class that I’ve started is Harvard Stats 110. Joe Blitzstein is lively and provides a lot of intuition in his classes.

I’m gathering all of my resources in one Github repo

Finally, to help me better absorb the material, I’m planning on writing several blog posts on probability, statistics, and machine learning.


Voice as a User Interface is Almost Here

Talking to your computer to interact with it has been science fiction fantasy for over 50 years. It will soon be a reality.

Amazon made voice activated devices popular with its Echo devices . Google responded with the Google Home.

We now have an arm’s race between these two giants.  Amazon or Google should get us over the finish line.

Apple has Siri, of course, but it’s currently in third place in capabilities, and it does’t offer cheap $30 Siri devices. Nevertheless, look for Siri improvements too.

One concern with these devices that’s always mentioned is that of privacy. Your request, along with your voice, is transmitted to the cloud for processing. That’s where all the computational power and most recent machine learning algorithms live. Apple’s data is anonymous, which I believe slows their ability to improve Siri’s abilities. Alexa and Google, for example, can recognize between different people’s voices.

Here’s a review of the Google Home Mini:

Marques also does a great comparison of the voice assistants:

Finally, if you’d like to help crowd source an open source voice solution, the Mozilla Foundation has Mozilla Common Voice where it needs people to both verify and donate their voice. This open source data will be used to help improve voice recognition.