Research map / Label-efficient and robust learning / Knowledge distillation
Discrete Survival Knowledge Distillation for Competing Risks Analysis
Feiyang Deng, Lingfeng Luo, Di Wang, Qinmengge Li, Lingxuan Kong, Kevin He
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 |
|---|---|---|---|
| distill a matched competing-risks teacher into the competing-risks student (DiSKD) | discrete-time competing-risks student · SRTR post-COVID cohort | Ctd (Death): 0.5954 -> 0.6889 | Improved |
| distill an overall-event teacher into the competing-risks student | discrete-time competing-risks student · SRTR post-COVID cohort | Ctd (Graft): 0.5667 -> 0.6533 | Improved |
| use a fixed-horizon binary teacher instead of a matched competing-risks teacher (ablation) | DiSKD · SRTR post-COVID cohort | Ctd (Death): 0.6889 -> 0.6306 (Binary-2) | Hurt |
| use a fixed-horizon binary teacher instead of internal-only training | competing-risks student · SRTR post-COVID cohort | predictive deviance: 0.4749 -> 0.4702 (Binary-1) | Improved |
| raise distillation temperature from T=1 to T=2 (ablation) | DiSKD-C with good teacher · simulation, correlated competing risks | Ctd and predictive deviance | Improved |
| raise temperature further from T=3 to T=5 (ablation) | DiSKD-C · simulation, correlated competing risks | Ctd and predictive deviance | No clear effect |
| use cause-specific teacher (DiSKD-C) rather than overall-event teacher (DiSKD-O) when teacher is good (ablation) | DiSKD · simulation, teacher quality (a) Good | Ctd and predictive deviance | Improved |
| use cause-specific teacher rather than overall-event teacher when teacher is poor (ablation) | DiSKD · simulation, teacher quality (c) Poor | Ctd and predictive deviance | No clear effect |
| distill from both teachers (multi-teacher) vs weaker teacher alone (ablation) | DiSKD · simulation, two teachers with different covariates | predictive performance | Improved |
| add finetuning from teacher before DiSKD (Finetune+DiSKD vs DiSKD) (ablation) | DiSKD · SRTR post-COVID cohort | concordance and deviance | No clear effect |
| use DiSKD instead of stacking (pooled training) or finetuning | post-COVID model update · SRTR post-COVID cohort | concordance-deviance trade-off | Improved |
| distill post-COVID model into donor-only student | donor-only student · SRTR donor-quality, post-COVID test cohort | Ctd death and graft failure | Improved |
Closest papers by meaning
- Orthogonal Survival Learners for Estimating Heterogeneous Treatment Effects from Time-to-Event Data (NeurIPS 2025 · similarity 0.85)
- SurvivalPFN: Amortizing Survival Prediction via In-Context Bayesian Inference (NeurIPS 2026 · similarity 0.85)
- PLD: A Choice-Theoretic List-Wise Knowledge Distillation (NeurIPS 2025 · similarity 0.84)
- Deep Copula-Based Survival Analysis for Dependent Censoring with Identifiability Guarantees (AAAI 2024 · similarity 0.84)
- Toward Conditional Distribution Calibration in Survival Prediction (NeurIPS 2024 · similarity 0.84)
- Domain constraints improve risk prediction when outcome data is missing (ICLR 2024 · 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.
