VAE: The KL to the Prior
Intermediate
The problem, from first principles
The VAE objective includes a KL term that pulls each posterior toward a standard normal prior. This keeps the latent space smooth, so random samples from the prior decode to plausible data.
From theory to code
Implement kl_std_normal(mu, logvar), the KL from a diagonal Gaussian to the standard normal, summed over dimensions.
Constraints
- Sum over dimensions and return a Python float.
Hints
Hint 1
Use -0.5 * sum(1 + logvar - mu^2 - exp(logvar)).
Keywords
research-papersunsupervisedvaelatent-variables
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.