AI Paper Atlas

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

A Probabilistic Hard Concept Bottleneck for Steerable Generative Models

María Martínez-García, Ricardo Vazquez Alvarez, Alejandro Lancho, Pablo M. Olmos, Isabel Valera

ICLR 2026 · Probabilistic and scientific modeling › Variational and generative sampling · approach: Disentangled and discrete latent 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 VHCB hard probabilistic concept bottleneck instead of CB-AEStyleGAN2 CelebA-HQ · single intervention inactive to active, low corr., ResNet18target accuracy: 0.42 -> 0.77Improved
use VHCB instead of CB-AEStyleGAN2 CelebA-HQ · single intervention, non-targetnon-target accuracy: 0.83 -> 0.76Hurt
use VHCB instead of CB-AEStyleGAN2 CelebA-HQ · concept inference, all concepts, CLIP labelsaccuracy: 0.565 -> 0.623Improved
use VHCB instead of CB-AEStyleGAN2 CelebA-HQ · disentanglement of c and s, low corr., ResNet18accuracy: 0.701 -> 0.874Improved
sample concept patterns by training frequency instead of random concept setsVHCB · direct generation, all concepts, ResNet18target accuracy: 0.551 -> 0.873Improved
target minimum Hamming distance training patterns instead of single arbitrary conceptVHCB · interventions, all concepts, ResNet18, active to inactivetarget accuracy: 0.550 -> 0.660Improved
use a low-correlation concept set instead of all 40 conceptsVHCB · single intervention inactive to active, ResNet18target accuracy: 0.170 -> 0.769Improved
add unsupervised side channel s (ablation)VHCB · CelebA-HQ balanced concepts, FIDFID: 20.950 -> 11.016Improved
add unsupervised side channel s (ablation)VHCB · CelebA-HQ all concepts, FIDFID: 7.921 -> 7.248No clear effect
use symmetric KL instead of BCE concept loss (ablation)VHCB · CelebA-HQ balanced, ResNet18, a to itarget accuracy: 0.564 -> 0.814Mixed
use symmetric KL instead of BCE concept loss (ablation)VHCB · CelebA-HQ balanced, ResNet18, i to atarget accuracy: 0.395 -> 0.375Hurt
increase side channel KL weight from 0 to 40 (ablation)VHCB · CelebA-HQ balanced, s size 5, generationtarget accuracy: 0.659 -> 0.561Hurt

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