Research map / Time series / Causal and generative time series
Not to Cry Wolf: Distantly Supervised Multitask Learning in Critical Care
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 |
|---|---|---|---|
| use 12 auxiliary tasks selected by feature importance | DSMT · ICU alarm classification, 25 labels | AUROC: 0.834 -> 0.872 | Helped |
| use 50 auxiliary tasks selected by feature importance | DSMT · ICU alarm classification, 50 labels | AUROC: 0.834 -> 0.901 | Helped |
| use the proposed DSMT approach | ICU alarm classification · 25, 50 and 100 labels | AUROC: 13.0%, 12.6% and 8.0% over Ladder Networks; 9.4%, 10.4% and 5.6% over Feature Matching GANs | Helped |
| remove two-step training; instead train auxiliary and main tasks without the alternating procedure (ablation) | DSMT-100 · ICU alarm classification, more than 12 labels | AUROC: 0.831 -> 0.798; 0.893 -> 0.785; 0.907 -> 0.814; 0.934 -> 0.849; 0.934 -> 0.898 | Hurt |
| add two-step training that alternates auxiliary-task and main-task optimization (ablation) | DSMT-100 · ICU alarm classification, more than 12 labels | AUROC: 0.798 -> 0.831; 0.785 -> 0.893; 0.814 -> 0.907; 0.849 -> 0.934; 0.898 -> 0.934 | Helped |
| distribute auxiliary tasks sequentially in depth rather than in parallel (ablation) | DSMT-100 · ICU alarm classification | AUROC: 0.720 vs 0.587; 0.831 vs 0.611; 0.893 vs 0.722; 0.907 vs 0.610; 0.934 vs 0.624; 0.934 vs 0.702 | Hurt |
| place auxiliary tasks directly on the head block (ablation) | DSMT · ICU alarm classification | AUROC: 0.791 vs 0.720; 0.804 vs 0.831; 0.828 vs 0.893; 0.887 vs 0.907; 0.941 vs 0.934; 0.940 vs 0.934 | Mixed |
| select 6 auxiliary tasks randomly rather than by feature-importance ranking (ablation) | DSMT-6 · ICU alarm classification | AUROC: 0.763 -> 0.805; 0.839 -> 0.851; 0.866 -> 0.884; 0.897 -> 0.909 | Helped |
| select 100 auxiliary tasks randomly rather than by feature-importance ranking (ablation) | DSMT-100 · ICU alarm classification | AUROC: 0.720 -> 0.790; 0.831 -> 0.860; 0.893 -> 0.883; 0.907 -> 0.909; 0.934 -> 0.918; 0.934 -> 0.932 | Mixed |
| increase the number of feature-importance-selected auxiliary tasks (ablation) | DSMT · ICU alarm classification with sufficient labels | AUROC | Helped |
| increase the number of randomly selected auxiliary tasks from 6 to 100 (ablation) | DSMT · ICU alarm classification | AUROC | No clear effect |
| use Feature Matching GAN instead of DSMT | ICU alarm classification · 12 labels | AUROC: 0.846 vs 0.763 | Hurt |
Closest papers by meaning
- GARLIC: Graph Attention-based Relational Learning of Multivariate Time Series in Intensive Care (ICLR 2026 · similarity 0.88)
- Learning Implicit Tasks for Patient-Specific Risk Modeling in ICU (AAAI 2017 · similarity 0.86)
- Yet Another ICU Benchmark: A Flexible Multi-Center Framework for Clinical ML (ICLR 2024 · similarity 0.85)
- CLIMB: Data Foundations for Large Scale Multimodal Clinical Foundation Models (ICML 2025 · similarity 0.83)
- XTSFormer: Cross-Temporal-Scale Transformer for Irregular-Time Event Prediction in Clinical Applications (AAAI 2025 · similarity 0.82)
- MAESTRO : Adaptive Sparse Attention and Robust Learning for Multimodal Dynamic Time Series (NeurIPS 2025 · similarity 0.82)
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.