AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Formalizing Consistency and Coherence of Representation Learning

Harald Stromfelt, Luke Dickens, Artur Garcez, Alessandra Russo

NeurIPS 2022 · Label-efficient and robust learning › Recognition and long-tail · approach: Contrastive representation theory · 0 citations (OpenAlex)

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 Dynamic Comparator relation-decoder instead of HolErelation-decoder · PRT MNIST->BlockStacksaggregate coherence (consistency loss): 12.12 -> 0.53Improved
use Dynamic Comparator instead of NNrelation-decoder · PRT MNIST->BlockStacksaggregate coherence (consistency loss): 1.71 -> 0.53Improved
use Dynamic Comparator instead of TransRrelation-decoder · PRT MNIST->BlockStacksaggregate coherence (consistency loss): 2.51 -> 0.53Improved
use Dynamic Comparator instead of NTNrelation-decoder · PRT MNIST->BlockStacksCon-A coherence (consistency loss): 1.92 -> 0.29Improved
use Dynamic Comparator instead of NN on reflexivityrelation-decoder · PRT MNIST->BlockStacksCon-I-R coherence: 0.05 -> 0.09Mixed
increase disentanglement pressure beta in {1,4,8,12} (ablation)DC · PRT MNIST->BlockStackstarget-domain accuracy of held-out relationsNo clear effect
increase beta (ablation)NN, NTN, HolE · PRT MNIST->BlockStackstarget-domain PRT accuracyMixed
increase beta (ablation)TransR · PRT MNIST->BlockStackstarget-domain PRT accuracyImproved
increase beta (ablation)all models · PRT MNIST->BlockStackssource-domain accuracyNo clear effect
transfer only the isSuccessor decoder to the target domainNN, NTN, HolE · PRT MNIST->BlockStacksaccuracy on held-out relationsHurt
evaluate consistency on interpolated latent samples instead of data embeddingsconsistency losses · PRT MNIST->BlockStacksSpearman correlation with PRT accuracyImproved
increase beta (ablation)NN, NTN · PRT MNIST->BlockStacksCon-I consistency lossesImproved

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