Research map / Architectures and efficiency / RNNs and normalization
RETAIN: An Interpretable Predictive Model for Healthcare using Reverse Time Attention Mechanism
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| temporal RNN models instead of stationary models | LR/MLP · Heart failure prediction (Sutter) | AUC: 0.7900 -> 0.8706 | Helped |
| MLP instead of LR | aggregated-feature model · Heart failure prediction (Sutter) | AUC: 0.7900 -> 0.8256 | Helped |
| RETAIN two-level attention vs plain RNN | RNN · Heart failure prediction (Sutter) | AUC: 0.8706 -> 0.8705 | No clear effect |
| RETAIN vs plain RNN | RNN · Heart failure prediction (Sutter) | test negative log-likelihood: 0.2577 -> 0.2562 | Helped |
| scalar visit attention generated by RNN (alpha_R) instead of MLP (alpha_M) (ablation) | RNN+attention · Heart failure prediction (Sutter) | AUC: 0.8624 -> 0.8717 | Helped |
| add vector-level beta attention to alpha_R (RETAIN) (ablation) | RNN+alpha_R · Heart failure prediction (Sutter) | AUC: 0.8717 -> 0.8705 | No clear effect |
| add beta attention to alpha_R (RETAIN) (ablation) | RNN+alpha_R · Heart failure prediction (Sutter) | test negative log-likelihood: 0.2605 -> 0.2562 | Helped |
| RETAIN vs RNN training cost | RNN · Heart failure prediction (Sutter) | train time per epoch: 10.3s -> 10.8s | Hurt |
| RETAIN vs plain RNN | RNN · Encounter diagnosis prediction | Recall@10: 69.09 -> 67.74 | Hurt |
| RETAIN vs plain RNN | RNN · Encounter diagnosis prediction | Recall@5: 55.18 -> 54.25 | Hurt |
| RETAIN vs MLP | MLP · Encounter diagnosis prediction | Recall@10: 65.02 -> 67.74 | Helped |
| MLP vs RNN+alpha_M (stationary vs sequential) | MLP · Encounter diagnosis prediction | Recall@10: 65.02 -> 65.81 | No clear effect |
Closest papers by meaning
- Context Clues: Evaluating Long Context Models for Clinical Prediction Tasks on EHR Data (ICLR 2025 · similarity 0.85)
- One Loss to Rule Them All: Marked Time-to-Event for Structured EHR Foundation Models (NeurIPS 2026 · similarity 0.85)
- MOTOR: A Time-to-Event Foundation Model For Structured Medical Records (ICLR 2024 · similarity 0.85)
- Toward Better EHR Reasoning in LLMs: Reinforcement Learning with Expert Attention Guidance (AAAI 2026 · similarity 0.84)
- SMART: Towards Pre-trained Missing-Aware Model for Patient Health Status Prediction (NeurIPS 2024 · similarity 0.83)
- Contrastive Learning for Clinical Outcome Prediction with Partial Data Sources (ICML 2024 · similarity 0.83)
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.