Problems with basic GANs

Lack of synchronisation, mode collapse and unstable training.

The generator is trained against the previous discriminator, while the discriminator is trained on samples from the previous generator — the two can fall out of sync.

Basic GANs also suffer from mode collapse (the generator produces only a few kinds of output) and unstable training. IPM-based GANs can provide more stable training with stronger convergence.

Formula to remember

RBM energy

E (v, h) = - a^{⊤} v - b^{⊤} h - v^{⊤} W h

Visible–hidden interactions only.

Quick check · +5 XP

An RBM differs from a general Boltzmann machine because it has…

Practice this topic

Your notes

Saved in this browser as you type.