AI Paper Atlas

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

A Differential Entropy Estimator for Training Neural Networks

Georg Pichler, Pierre Jean A. Colombo, Malik Boudiaf, Günther Koliander, Pablo Piantanida

ICML 2022 · Probabilistic and scientific modeling › Variational and generative sampling · approach: Mutual information estimation · 1 citations (OpenAlex)

Read the paper

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 KNIFE-based MI estimation instead of competing MI estimators or VIBERT in the information bottleneck objectiveBERT fine-tuning · GLUE: MRPC, STS-B, RTEtask scores: MRPC F1 85.1 -> 85.3; MRPC accuracy 89.1 -> 90.1; STS-B Pearson 90.0 -> 90.3; STS-B Spearman 89.7 -> 90.1; RTE accuracy Improved
Use KNIFE-based MI estimation rather than the competing estimatorsUnsupervised domain adaptation · Mean across 6 source/target tasksaccuracy: 82.8 -> 83.4Improved
Use KNIFE-based MI estimation rather than other MI estimatorsFair textual classification · DIAL sentiment and mention labelsmain-task accuracy at perfect privacyMixed
Allow kernel means and mixture weights to be learned, rather than fixing kernel supportDifferential entropy estimator · Changing input distributionadaptation to distribution shiftImproved
Use a flexible kernel mixture rather than a single Gaussian densityDifferential entropy estimator · Multimodal triangle-mixture distributionsdensity matchingImproved

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