TPE-AS was designed around a controlled evaluator that isolates private models, data, and backtesting infrastructure. The public artifact focuses on reusable optimizer mechanics and synthetic black-box benchmarks.
← All work
TPE-AS
Adaptive Bayesian optimization for expensive, noisy portfolio-model tuning.
Problem
Private portfolio models expose mixed, expensive, noisy objectives with limited evaluation budgets and no gradient access.
Approach
Add a budget-dependent mean-variance objective and clipped importance correction to good/bad Parzen-density search.
Role
Lead researcher and implementation author