Gate: rescue a GAN that learns nothing (AI off)
Why it matters A GAN that fails rarely fails for one reason. Here three settings are wrong at once; some matter and some only matter in company. Finding which is the skill.
Gate rules: AI tutor off. The broken GAN uses 8 noise numbers, the saturating loss and a slow generator (g_lr 2.5e-4). The starter shows its recorded runs. Train variations with ganlab.train(seed=0, ...) and judge them with ganlab.coverage; each run takes about 4 seconds, so try one or two per run.
The check trains your fix on a new seed and passes it with at least 7 blobs covered and quality at least 0.5.
When you are done, ask yourself which of the three settings you changed were really needed. The reveal after the check has the measured answer.
Words and symbols in this lesson
- GAN
- generative adversarial network: a generator and a discriminator trained against each other
- generator (G)
- the network that turns noise into a sample
- saturating loss
- G minimises log(1 - D(G(z))); its slope is about 0 once D is sure, so G stops learning
- blobs covered, quality
- the ring judge: blobs with more than 2% of the points, and the share of points on a blob
Set fix to a dict of settings for ganlab.train (any of noise, loss, g_lr, d_lr, d_steps, smooth, instance_noise, spectral).
AI off: no tutor, hints or worked solution on this one. Your own thinking only.
Spotted a problem in this lesson? Report it · Ask about this lesson
This exercise opens soon
Finish the lesson before it and this one unlocks. The whole course is free, no account needed, and you can read this lesson right now.
This browser is blocking cookies for learn.floati.life, so lessons you finished here cannot open in order. Allow cookies for this site and reload.