AI Paper Atlas

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

ELPAC: Endpoint-Anchored Latent Progression with Stage-Varying Multimodal Coordination

Shengxian DING, Yifei Zhang, Xinyuan Tian, Yuanshuang Guo, Zhao

NeurIPS 2026 · Probabilistic and scientific modeling › Variational and generative sampling · approach: Hierarchical VAEs and posterior collapse · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes ELPAC, a method for jointly inferring latent disease stage, progression subtype, and stage-varying multimodal coordination from partially observed cross-sectional aging cohort data (imaging, plasma, cognition).

Problem

Aging cohorts with imaging, plasma, and cognitive measurements usually give only partially observed cross-sectional snapshots, so latent stage, progression subtype, and how modalities coordinate across stages must be inferred jointly.

Approach

Endpoint-anchored latent progression model with stage-varying multimodal coordination, as described by the title; the abstract is truncated so the precise mechanism is not stated.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add endpoint anchors (ablation)ELPAC · Synthetic cohorts (N=1,000)subtype BAcc: 48.9 -> 95.4Improved
add endpoint anchors (ablation)ELPAC · Synthetic cohorts (N=1,000)stage Spearman rho: 92.4 -> 96.2No clear effect
add endpoint anchors (ablation)ELPAC · Synthetic cohorts (N=1,000)|Cog r|: 76.1 -> 97.9Improved
use factor covariance instead of diagonal (ablation)ELPAC · Synthetic cohorts (N=1,000)held-out NLL: 23.97 -> 19.27Improved
use factor covariance instead of diagonal (ablation)ELPAC · Synthetic cohorts (N=1,000)subtype BAcc: 68.7 -> 95.4Improved
make factor activations stage-varying instead of static (ablation)ELPAC · Synthetic cohorts (N=1,000)stage-binned PPC: 78.7 -> 82.8Improved
make factor activations stage-varying instead of static (ablation)ELPAC · Synthetic cohorts (N=1,000)subtype BAcc: 95.5 -> 95.4No clear effect
add endpoint anchors (ablation)ELPAC · ADNI+NACC/SCAN cohortsubtype BAcc: 64.2 -> 68.5Improved
use stage-varying factors instead of static (ablation)ELPAC · ADNI+NACC/SCAN cohortsubtype BAcc: 58.7 -> 68.5Improved
use factor covariance instead of diagonal (ablation)ELPAC · ADNI+NACC/SCAN cohortheld-out NLL: 12.83 -> 11.01Improved
add supervised classification head (HEALNet) vs ELPACHEALNet · ADNI+NACC/SCAN cohortAUC AD/(CN+MCI): 86.4 -> 87.8Hurt
increase anchor mislabeling to 20% (ablation)ELPAC · Synthetic and real cohortsBAcc / stage rho: 3.4 and 5.7 pp drops; real BAcc 7.0 ppHurt

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