20 Apr 2024

Modern AI models such as OpenAI’s GPT, Anthropic’s Claude, Meta’s LLaMA, and Google’s Gemini have set off a boom in data center construction and a staggering rally in Nvidia thanks to the demand for their (over USD40,000) chips; machine learning models suck up a lot of electricity due to their compute-expensive training loops, and total ML modeling power consumption is as much as a small country. Small wonder that there is interest in energy-efficient machine learning methods; from the last article (emphasis mine):

02 Apr 2024

This is a quick, no-frills introduction to writing GPU code on Kaggle, which offers free notebooks for analysis thanks to a tie-up with Google. Kaggle notebooks are similar to Colab instances, but they also offer cheaper and higher memory GPU-enabled instances (as of this writing).

26 Dec 2021

This post is a bit of a deviation away from my usual posts insofar as it avoids the pleasantries of my previous introductions and conclusions. Rather, this is a quick note about the Elo rating system, invented by the physicist Arpad Elo (frequently referred to as the ELO rating system, under the assumption that ‘Elo’ was an acronym rather than the surname of the inventor).

26 Nov 2021

In 1972, the economist Kenneth Arrow received the Nobel Memorial Prize in Economics for a profound discovery published as part of his 1951 doctoral dissertation. In it, he demonstrated an *impossibility theorem*: a mathematical result showing that certain conditions were impossible. In this case, it was a discovery about voting systems which proved decisive in establishing the field of social choice theory.

26 May 2021

What does naming big numbers have to do with the technological sophistication of a society?

29 Sep 2020

As we enter the second wave of the coronavirus pandemic and fix our gaze upon political spectacles with the lucidity of a fever dream, we might ruminate on the path that led to our present state. Modernity seems to lurch from crisis to crisis every year — from financial instability (the “great recession” of 2008-2009) to nuclear disasters (Fukushima) to zoonotic pandemics (COVID-19).

20 Jun 2020

P-values are often mentioned in the scientific literature, but what exactly are they? Short for *probability-value*, they represent tail-area probabilities of particular distributions and are primarily used in *hypothesis testing* (for frequentists) and in *model checking* (for bayesians). In this short post I’ll try to elucidate the two main philosophies on using p-values along with some common pitfalls — the dreaded *p-hacking*, for instance, along with some pointers on corrections for repeated testing.

09 Apr 2020

Let’s say you want to model binary outcomes. This is simple if you want a maximum likelihood estimate of a linear set of weights — just perform a logistic regression. But what do we do if we want to perform Bayesian inference?

01 Mar 2020

This blog is intended as a collection of gleanings and sowings — récoltes et semailles — of morsels of knowledge accumulated through some years of trial and effort in my attempts to understand topics that I found engaging or difficult.