Recurrent neural networks are one of the staples of deep learning, allowing neural networks to work with sequences of data like text, audio and video. They can be used to boil a sequence down into a high-level understanding, to annotate sequences, and even to generate new sequences from scratch!

Back when I was a kid in England, I used to dabble with electronics. I’d solder little projects together and wire them up with cables of various thicknesses. The diameters of the hook-up wires I’d use (and the diameter of the solder wire) were measured in units called SWG (Standard Wire Gauge). |Back when I was a kid in England, I used to dabble with electronics. I’d solder little projects together and wire them up with cables of various thicknesses. The diameters of the hook-up wires I’d use (and the diameter of the solder wire) were measured in units called SWG (Standard Wire Gauge).In what was perverse logic to a child; thin wires were given large SWG numbers, and thicker wires were given smaller SWG numbers.You might have noticed that drill sizes also seem to follow this same inverted standard: Large numbers for small diameter drills, and small numbers for large diameter drills.|

This guide is patterned after my “Doing well in your courses”, a post I wrote a long time ago on some of the tips/tricks I’ve developed during my undergrad. I’ve received nice comments about that guide, so in the same spirit, now that my PhD has come to an end I wanted to compile a similar retrospective document in hopes that it might be helpful to some. Unlike the undergraduate guide, this one was much more difficult to write because there is significantly more variation in how one can traverse the PhD experience. Therefore, many things are likely contentious and a good fraction will be specific to what I’m familiar with (Computer Science / Machine Learning / Computer Vision research). But disclaimers are boring, lets get to it!

As machine learning increasingly affects domains protected by anti-discrimination law, there is much interest in the problem of algorithmically measuring and ensuring fairness in machine learning. Across academia and industry, experts are finally embracing this important research direction that has long been marred by sensationalist clickbait overshadowing scientific efforts.

Very often one builds a statistical model in pieces. For example, imagine one has a binary event which may or may not occur - to work with my thematic example, a visitor arrives on a webpage and he may or may not convert. A reasonable question to ask is “if I have 100 visitors, how many of them can I expect to convert?” Assume now that I know the conversion rate lmbda; in this case the maximum likelihood point estimate for the number of conversions is 100*lmbda and the probability distribution of possible events that could occur is binom(100, lmbda) (i.e. a binomial distribution). But what happens if lmbda is not known, but instead a random variable?

Back in 2014, the U.S. Department of Transportation released a report on the (normalized) number of traffic deaths in each U.S. state. As I looked through the list, I noticed an odd correlation between the political leanings of a state and its traffic fatalities.

The Python 2 vs. Python 3 divide has long been a thorn in the Python community’s side. On one hand, Python package developers face the challenge of supporting two incompatible versions of Python, which is time that could be better spent improving the package. On the other hand, many Python users are reluctant to upgrade from Python 2 to 3 because of the time commitment such an upgrade entails. The Python Software Foundation’s official stance on the matter is:

Profile Service

Tricks of the optimization trade¶