All Research

A computational model simulating the delivery of therapeutic agents across the blood-brain barrier. This project explores novel methods for increasing drug efficacy for neurological disorders.

This repository develops and validates an infinite-dimensional Bayesian posterior operator for updating distributions over asset weights or model parameters. It supports nonparametric modeling using Gaussian process priors to fully quantify uncertainty in hierarchical asset allocation.

This module introduces a novel regularization framework for portfolio optimization that explicitly penalizes allocation strategies based on their Kolmogorov complexity, aiming to favor simpler, more robust strategies.

This module presents a rigorous formulation of a Multi-Information Bottleneck (MIB), designed to compress multiple predictive signals into a robust, minimal representation for portfolio decision-making.

This repository provides a reproducible implementation of a duality-informed convex equilibrium allocation theory. It features a robust solver that finds portfolio weights by minimizing a convex objective that combines risk, return, and distributional robustness, while enforcing simplex constraints.

This module provides a production-ready implementation for a CVaR-Adjusted Update to portfolio weights. The method explicitly penalizes tail risk by down-weighting assets with larger Conditional Value-at-Risk (CVaR), making allocations more robust to extreme downside events.

This module provides a production-ready implementation of a portfolio weight adjustment mechanism inspired by reinforcement learning. It updates weights using a simple multiplicative rule based on a reward signal, ensuring the portfolio remains valid and normalized.

A production-ready implementation of an Entropy Regularization Module. This module encourages portfolio diversification and robustness by penalizing concentrated allocations, using an entropy term and an exponentiated-gradient update.

This module provides a production-ready implementation of an autocorrelation penalty for portfolio updates. The method detects serial correlation in a signal using a Ljung-Box test and applies a shrinkage penalty to reduce exposure when strong autocorrelation is present.

A production-ready implementation of a Wasserstein Distributionally Robust Optimization (DRO) penalty for portfolio updates. This module hedges against distribution shift by penalizing the divergence between historical mean returns and a stressed or perturbed mean.