Mini-batch Estimation for Deep Cox Models: Statistical Foundations

Lang Zeng's research provides statistical foundations and practical guidance for using stochastic gradient descent (SGD) to optimize deep Cox neural networks. The work establishes the consistency and convergence rate of the mini-batch maximum partial-likelihood estimator (mb-MPLE) and demonstrates its effectiveness in large-scale applications where standard methods are intractable. It offers insights into hyperparameter tuning, particularly the critical ratio of learning rate to batch size.

Use Cases

• Hyperparameter tuning for SGD in Cox-NN based on the learning rate to batch size ratio
• Approximating the global optimizer (mb-MPLE) for Cox regression using iterative convergence analysis
• Applying mb-MPLE for survival analysis in large-scale real-world applications where standard MPLE is intractable
• Simulating and validating the asymptotic normality and variance properties of mb-MPLE for linear Cox models

Strengths

• Dataset includes a 3.2 MB file containing supporting materials for the research
• Provides theoretical analysis and numerical evidence for practical guidance on using SGD
• License is CC-BY-4.0, allowing for open sharing and reuse

Limitations

• Column names and row counts are not provided, limiting understanding of the specific data structure
• The dataset appears to be simulation outputs or code rather than a primary observational dataset
• The description focuses on methodological results, not on the raw data content

Provenance

Source

Lang Zeng

Collection Method

Likely generated through simulation studies and theoretical analysis as part of academic research.

Freshness

Last updated on 2026-03-20