Top 5 algorithms that changed the world

12 November, 2019
Episode notes

Guest

Dr Sophie Carr

World’s most interesting mathematician and #Queenof Statistics winner. Owner of Bays Consulting. Loves Bayesian statistics.


Mark gets mathematical with bayesian mathematician Sophie Carr.

Sophie describes herself as an “accidental mathematician”, getting into the subject after studying to be an engineer. She got her PhD on the job, and transitioned from fluid mechanics to Bayesian statistics.

Here’s the piping speech from Patriot Mark mentioned.

https://www.youtube.com/watch?v=P5-9Rfrui9A

Sophie’s picks

In order of discussion:

Turbo Code

Invented by telephone engineer Claude Berrou, this bayesian algorithm encodes information for use in phone calls, and has laid the way for technology like 3G and 4G, a couple of decades in advance.

Universal unique identifier (UUID)

UUIDS — sometimes galled GUIDs — are collections of numbers and letters that can be generated by independent machines, and be almost guaranteed to be entirely unique. They’re based on timestamps — a number of seconds since a given period — and some info about the computer creating the code, as well as some random bits.

PageRank

Named after Google co-founder Larry Page, the PageRank algorithm ranks web pages based on links, the assumption being that the more important a website is, the more people will link to it.

Proportional integral derivative control

A PID controller corrects problems with automated processes. The best example, and why Sophie added it to her list, is ABS (automatic breaking in a car).

Mark’s picks

In order of discussion:

LZ 77 compression

This simple form of text compression is Mark’s top pick because it forms part of the Zip archive format, which is something even non-technical people are familiar with, and have used for decades. It works by dropping markers to repeated characters within a file, so you could reduce the phrase “how now brown cow” to “h① n① br①n c①” (a reduction of around 20%).

RSA

Public key — or asymmetric — cryptography allows for information to be encrypted by one party, but only readable by the other party. That works because the party that wants to receive the information makes their public key available to anyone who wants it, and then those that want to send information, use that public key to encrypt it. However the information can only be decrypted by that first person using their private key.

Proof of Work

This algorithm is central to the block chain – which is what bitcoin and other cryptocurrencies are based on, but is also how it’s possible to maintain a shared ledger to reliably keep track of information, and to keep people honest.

Reinforcement learning

This machine learning algorithm made it onto MArk’s list basically because it’s nice. It reinforces what a computer has learned — hence the name — by giving it a reward; it’s the computer’s job to maximise the time-to-reward ratio, so will figure out ways to solve the problem more efficiently to get that sweet, sweet reward.

Dynamic range compression

Mark uses this audio compression algorithm in his work as a podcast producer. Essentially it replicates the job of a human, turning the volume of a speaker or an instrument so that the quiet parts are audible, but the loud parts don’t blow your ears off.

More of Sophie

Sophie runs Bays Consulting, who you can follow on Twitter along with Sophie herself.

Links