probml / pml2-book

Probabilistic Machine Learning: Advanced Topics
MIT License
1.39k stars 120 forks source link

Version 2023-08-15, 26.2.2: minimization vs maximization of the Jensen-Shannon divergence #310

Closed tstafylakis closed 3 months ago

tstafylakis commented 7 months ago

26.2.2 Density ratio estimation using binary classifiers: The authors described V as "minimization of the Jensen-Shannon divergence". However, they refer to eq. 26.10 which optimizes the objective with respect to the parameters of the critic/discriminator. Therefore it should be maximization of the JS divergence.

murphyk commented 3 months ago

fixed, thx