Six (and a half) intuitions for KL divergence - LessWrong
lesswrong.com · 7,569 words · saved by 15 readers
KL-divergence is a topic which crops up in a ton of different places in information theory and machine learning, so it's important to understand well…
x Six (and a half) intuitions for KL divergence — LessWrong Information theory Probability & Statistics Machine Learning (ML) World Modeling Frontpage 185 Six (and a half) intuitions for KL divergence by CallumMcDougall 12th Oct 2022 Linkpost for www.perfectlynormal.co.uk 12 min read 27 185 KL-divergence is a topic which crops up in a ton of different places in information theory and machine learning, so it's important to understand well. Unfortunately, it has some properties which seem confusing at a first pass (e.g. it isn't symmetric like we would expect from most distance measures, and it
saved by
- Kevin Wang
- Karan MJ
- Claire Wang
- Derek Yen
- Asher P
- Sarah Pan
- Can Rager
- Lydia Nottingham
- Swaminathan Gurumurthy
- Yudhister Joel Kumar
- Timothy Kostolansky
- Dhruv Sheth
related reading
- Approximating KL Divergencejoschu.net
- Kullback–Leibler divergence - Wikipediaen.wikipedia.org
- Visual Information Theory -- colah's blogcolah.github.io
- blog.alexalemi.com Why KL?blog.alexalemi.com
- Andy Jonesandrewcharlesjones.github.io
- interpreting GPT: the logit lens — LessWronglesswrong.com
- Short Notes on Divergence Measuresdanilorezende.com
- Exponential families from a single KL identityarxiv.org
- Things that confused me about cross-entropychris-said.io
- blog.alexalemi.com KL is All You Needblog.alexalemi.com
- How much do you believe your results? — LessWronglesswrong.com
- [2206.07867] A visual introduction to information theoryarxiv.org