Publications by Fabian Dablander
Curve fitting and the Gaussian distribution
Judea Pearl said that much of machine learning is just curve fitting1 — but it is quite impressive how far you can get with that, isn’t it? In this blog post, we will look at the mother of all curve fitting problems: fitting a straight line to a number of points. In doing so, we will engage in some statistical detective work and discover the ...
20806 sym R (545 sym/1 pcs) 22 img
Variable selection using Gibbs sampling
“Which variables are important?” is a key question in science and statistics. In this blog post, I focus on linear models and discuss a Bayesian solution to this problem using spike-and-slab priors and the Gibbs sampler, a computational method to sample from a joint distribution using only conditional distributions. Variable selection is a be...
41343 sym R (9460 sym/16 pcs) 18 img
Two perspectives on regularization
Regularization is the process of adding information to an estimation problem so as to avoid extreme estimates. Put differently, it safeguards against foolishness. Both Bayesian and frequentist methods can incorporate prior information which leads to regularized estimates, but they do so in different ways. In this blog post, I illustrate these two...
18106 sym R (1859 sym/3 pcs) 10 img
Bayesian modeling using Stan: A case study
Practice makes better. And faster. But what exactly is the relation between practice and reaction time? In this blog post, we will focus on two contenders: the power law and exponential function. We will implement these models in Stan and extend them to account for learning plateaus and the fact that, with increased practice, not only the mean re...
27853 sym R (9406 sym/40 pcs) 26 img
Spurious correlations and random walks
The number of storks and the number of human babies delivered are positively correlated (Matthews, 2000). This is a classic example of a spurious correlation which has a causal explanation: a third variable, say economic development, is likely to cause both an increase in storks and an increase in the number of human babies, hence the correlation...
9514 sym R (2869 sym/10 pcs) 10 img
The Fibonacci sequence and linear algebra
Leonardo Bonacci, better known as Fibonacci, has influenced our lives profoundly. At the beginning of the $13^{th}$ century, he introduced the Hindu-Arabic numeral system to Europe. Instead of the Roman numbers, where I stands for one, V for five, X for ten, and so on, the Hindu-Arabic numeral system uses position to index magnitude. This leads t...
14507 sym R (1326 sym/13 pcs) 6 img
Love affairs and linear differential equations
Differential equations are a powerful tool for modeling how systems change over time, but they can be a little hard to get into. Love, on the other hand, is humanity’s perennial topic; some even claim it is all you need. In this blog post — inspired by Strogatz (1988, 2015) — I will introduce linear differential equations as a means to stud...
30871 sym R (2648 sym/12 pcs) 26 img
Harry Potter and the Power of Bayesian Constrained Inference
If you are reading this, you are probably a Ravenclaw. Or a Hufflepuff. Certainly not a Slytherin … but maybe a Gryffindor? In this blog post, we let three subjective Bayesians predict the outcome of ten coin flips. We will derive prior predictions, evaluate their accuracy, and see how fortune favours the bold. We will also discover a neat tric...
14680 sym R (828 sym/15 pcs) 14 img
A brief primer on Variational Inference
Bayesian inference using Markov chain Monte Carlo methods can be notoriously slow. In this blog post, we reframe Bayesian inference as an optimization problem using variational inference, markedly speeding up computation. We derive the variational objective function, implement coordinate ascent mean-field variational inference for a simple linear...
17012 sym R (5404 sym/10 pcs) 2 img
An introduction to Causal inference
Causal inference goes beyond prediction by modeling the outcome of interventions and formalizing counterfactual reasoning. In this blog post, I provide an introduction to the graphical approach to causal inference in the tradition of Sewell Wright, Judea Pearl, and others. We first rehash the common adage that correlation is not causation. We the...
40533 sym R (1166 sym/4 pcs) 28 img