recons = tf.reduce_sum(p_x.log_prob(x),1)

ericjang / gumbel-softmax

categorical variational autoencoder using the Gumbel-Softmax estimator

MIT License

425 stars 101 forks source link

recons = tf.reduce_sum(p_x.log_prob(x),1) #13

Closed gitfourteen closed 3 years ago

gitfourteen commented 3 years ago

Can you explain what recons stands for? Please not use acronyms or shorthand without annotation. Thank you!

gitfourteen commented 3 years ago

Should be reconstruction loss mentioned in Auto-Encoding Variational Bayes; but found no reference to this jargon in neither paper on Gumbel Softmax