Research map / Vision-language and multimodal / Cross-modal learning
Electronic Health Record Analysis via Deep Poisson Factor Models
Ricardo Henao, James T. Lu, Joseph E. Lucas, Jeffrey Ferranti, Lawrence Carin
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use multiple layers (DPFM, DPFA, nHDP) instead of single-layer topic models | topic models · 20 News, RCV1, Wiki | held-out perplexity | Improved |
| DPFM instead of DPFA-SBN (128-64-32) | deep Poisson factor model · 20 News | held-out perplexity: 827 -> 780 | Improved |
| DPFM instead of LDA | topic model · RCV1 | held-out perplexity: 1179 -> 908 | Improved |
| double hidden units, add third layer or more samples to DPFM (ablation) | DPFM · 20 News, RCV1, Wiki | held-out perplexity | No clear effect |
| share information across modalities (Naive2 stacked) instead of one DPFM per modality | DPFM on EHR · EHR held-out perplexity, Code, 128-64-32 | held-out perplexity: 182.365 -> 180.806 | Improved |
| multi-modality DMPFM instead of Naive1 | EHR model · EHR held-out perplexity, Code, 128-64-32 | held-out perplexity: 182.365 -> 166.759 | Improved |
| increase layers from two to three (ablation) | DMPFM · EHR held-out perplexity, Code | held-out perplexity: 167.423 -> 166.759 | Improved |
| increase model size from 64-32 to 128-64 (ablation) | DMPFM · EHR held-out perplexity, Code | held-out perplexity: 194.260 -> 167.423 | Improved |
| DMPFM instead of Naive2 for classification | discriminative DPFM · T2DM EHR 13-outcome classification, all modalities, 128-64-32 | mean test AUC: 0.757 -> 0.784 | Improved |
| stack all modalities (Naive2) instead of Codes-only-like single modality Naive1 | discriminative DPFM · T2DM EHR 13-outcome classification, 128-64-32 | mean test AUC: 0.732 -> 0.757 | Improved |
| use Codes modality instead of Med or Lab only (ablation) | Naive1 · T2DM EHR 13-outcome classification, 128-64-32 | mean test AUC: 0.590 -> 0.732 | Improved |
| DMPFM instead of sparse logistic regression (LASSO) | classification of T2DM outcomes · 13 T2DM outcomes | test AUC | Mixed |
Closest papers by meaning
- Learning of Cluster-based Feature Importance for Electronic Health Record Time-series (ICML 2022 · similarity 0.87)
- Disentangling Latent Risk Pathways via Bayesian Hypergraph Inference (ICML 2026 · similarity 0.86)
- Exploiting Negative Samples: A Catalyst for Cohort Discovery in Healthcare Analytics (ICML 2024 · similarity 0.84)
- DrFuse: Learning Disentangled Representation for Clinical Multi-Modal Fusion with Missing Modality and Modal Inconsistency (AAAI 2024 · similarity 0.84)
- Unified Insights: Harnessing Multi-modal Data for Phenotype Imputation via View Decoupling (NeurIPS 2024 · similarity 0.84)
- Context Clues: Evaluating Long Context Models for Clinical Prediction Tasks on EHR Data (ICLR 2025 · similarity 0.84)
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.
