AI Paper Atlas

Research map / Probabilistic and scientific modeling / Variational and generative sampling

Mutual Information Constraints for Monte-Carlo Objectives to Prevent Posterior Collapse Especially in Language Modelling

Gábor Melis, András György, Phil Blunsom

JMLR 2022 · Probabilistic and scientific modeling › Variational and generative sampling · approach: Variational methods for sequence models

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use 16 samples instead of 1 with KL objectiveDReG · PTB test, hs=128perplexity: 92.5 -> 91.7Improved
use Renyi objective instead of KL with 16 samples (ablation)DReG · PTB test, hs=128perplexity: 91.7 -> 91.6Improved
use 16 samples instead of 1VIMCO · PTB test, hs=128perplexity: 94.6 -> 92.1Improved
use Renyi objective instead of KL with 16 samples (ablation)VIMCO · PTB test, hs=128perplexity: 92.1 -> 92.5Hurt
use 16 samples instead of 1 with KL objectiveDReG · PTB test, hs=256perplexity: 83.9 -> 82.8Improved
use Renyi objective instead of KL with 16 samples (ablation)DReG · PTB test, hs=256perplexity: 82.8 -> 82.2Improved
use 16 samples instead of 1VIMCO · PTB test, hs=256perplexity: 84.7 -> 83.4Improved
use Renyi objective instead of KL with 16 samples (ablation)VIMCO · PTB test, hs=256perplexity: 83.4 -> 83.1Improved
use multi-sample MI-constrained objectives to raise latent usageDReG, hs=128 · PTB testper-token KL: 0.047 -> 0.131Improved
use multi-sample objectives to avoid collapse with discrete latentsVIMCO, hs=128 · PTB testper-token KL: 0.000 -> 0.164Improved
use same dropout mask for all latent samples of a sentence (ablation)DReG/VIMCO multi-sample · PTB validationNLL vs latent usage Pareto curveImproved
replace dropout with L2 regularization (ablation)DReG/VIMCO multi-sample · PTB validationbest NLLHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy