New BTYD models leverage amortized variational inference for faster customer analysis

Researchers have developed a new family of Buy-'Til-You-Die (BTYD) models that move beyond the traditional Poisson process assumption. These new models utilize a Weibull renewal process and employ an amortized variational inference scheme for efficient parameter estimation. This approach significantly reduces computation time, fitting a dataset of 5 million retail customers in 8 minutes compared to the estimated 3-4 days required by current state-of-the-art methods, without sacrificing predictive performance or interpretability. The framework also facilitates the incorporation of covariates, as demonstrated on a dataset of 4 million political donors. AI

IMPACT Accelerates customer base analysis and modeling with improved computational efficiency.

RANK_REASON The cluster contains a research paper detailing new statistical models and inference methods. [lever_c_demoted from research: ic=1 ai=0.7]

Read on arXiv stat.ML →

• 2020 US General election
• 4 million political donors
• 5 million online retail customers
• amortized variational inference
• Bayesian inference
• Buy-'Til-You-Die (BTYD) models
• machine learning
• Poisson process
• Weibull renewal process

• paper
• model release
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →