The Motion Planning course is going faster than I expected. I completed 2 weeks within 5 days. Thats good I guess, since it means I might get to the Capstone project before I take a vacation to India.

I found it curious that a log-normal distribution fits both tweets and lines of code. Doing some digging, I found some literature on the subject of message lengths on the internet. This study finds that across languages and mediums, comment lengths follow a log-normal distribution quite closely. They propose a mechanism related to the Weber-Fechner law, which suggests a logarithmic scale in degrees of perception. It seems reasonable that lengths of code lines would respond to the same mechanism.

This post is just a quick note on some of the pitfalls we encounter when dealing with high-dimensional problems, even when working with something as simple as a Gaussian distribution.

Some time back, I was involved in a discussion with folks at an India-based software company. An important question was asked - why isn’t this company as productive (defined as revenue/employee) as it’s western competitors, and what can be done to change this situation? In this discussion, I put forward an unexpected thesis: if this company were profit maximizing, then it’s productivity should always be far lower than any western company. During the ensuing conversation, I came to realize that very few people were aware of the Cobb-Douglas model of productio, on which I was basing my counterintuitive conclusions.

How neural networks build up their understanding of images

A busy week. I finished my Aerial Robotics course! The next in the Specialization is Computational Motion Planning, which I am more excited about – mainly because the curriculum goes more towards my areas of expertise. Aerial Robotics was challenging primarily because I was doing a lot of physics/calculus which I had not attempted since a long time.

In this post we are going to develop an autocomplete component for large data sets using Tries Data Structure and Collaborating Filtering to choose best book titles suggestions to users. It is interesting to notice that both “Algorithm Structure” and “Machine Learning” are working together towards the final solution. Full code and working application are provided together with results.

A week back I had the pleasure to talk on machine learning at PyConUK 2017 in the inaugural PyDataCardiff track. Tim Vivian-Griffiths and colleagues did a wonderful job building our second PyData conference event in the UK. The PyConUK conference just keeps getting better – 700 folk, 5 tracks, a huge kids track and lots of sub-events. Pythontastic! Cat Lamin has a lovely write-up of the main conference.

This is great work coming from the Uber AI labs, especially by Eli Bingham and Noah Goodman for leading this effort among an excellent group. I’ve met with them in-person on numerous occasions to discuss the overall design and implementation details. Pyro touches on interesting aspects in PPL research: dynamic computational graphs, deep generative models, and programmable inference.

Hello all and welcome to Part 3 of building the JackBlockChain — JBC. Quick past intro, in Part 1 I coded and went over the top level math and requirements for a single node to mine its own blockchain; I create new blocks that have the valid information, save them to a folder, and then start mining a new block. Part 2 covered having multiple nodes and them having the ability to sync. If node 1 was doing the mining on its own and node 2 wanted to grab node 1’s blockchain, it can now do so.