My Take on Bias-Variance Decomposition!

  • 13 May 2022
  • 2 replies

Userlevel 2
Badge +2
  • Active Contributor
  • 9 replies

Bias-Variance Decomposition is one of the most important concepts in ML. It is helpful in understanding the performance of a machine learning algorithm and understanding the issue of overfitting. Here's my take on the same, hopefully it might be a good refresher for you too! #machinelearning #ml #bias #variance

2 replies

Userlevel 3
Badge +1

Thanks for this @Fazil!

I think that even conceptually appreciating models as “variables” as you do here can be a useful exercise in general (beyond of the context of your article).

It’s sometimes easy to forget that a given dataset is almost always a sample from (or “subset of”) a larger population, and that even “optimised” model parameters (which are conditional on this sample) will be uncertain relative to the “true” parameters (if such a thing can even be said to exist).

There’s uncertainty all the way down!

Userlevel 2
Badge +2

Thank you @jloxh  for your appreciation and adding your valuable points 😊