Researchers have developed new adaptation interfaces to improve the performance of tabular foundation models (TabFMs) in time-to-event prediction tasks. These interfaces address the challenges of handling censored data and event-time dynamics, building upon prior work by integrating TabFMs with established methods like CoxPH and DeepHit. The study evaluated various adaptation strategies, including temporal zero-shot reformulation, classification-based fine-tuning, and survival-head adaptation, across numerous datasets. Results indicate that while zero-shot inference is effective for smaller datasets, supervised adaptation becomes more advantageous as data size increases, with CoxPH generally providing the most reliable interface for larger datasets. AI
IMPACT Enhances the applicability of foundation models to complex survival analysis tasks in healthcare and other domains.
RANK_REASON The cluster contains a research paper detailing new methods for adapting foundation models. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- CoxPhLb: An R Package for Analyzing Length Biased Data under Cox Model
- DeepHIT: a deep learning framework for prediction of hERG-induced cardiotoxicity
- Hugging Face
- Integrated Brier Score
- Mannitol repressor MtlR-like
- tabular foundation models
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
