Selected work

Representative Papers

From Application-Layer Simulation to Native Meta-Architecture: Structural Tension as an Endogenous Driver for Heterogeneous AI Evolution

Jul 07, 2026

Current large language models lack intrinsic evolutionary capacity, relying instead on external prompt engineering that often leads to homogenization. This work proposes a native meta-architecture that, for the first time, leverages structural tension as an endogenous driving force, integrating offline recursive loops with inference-time plasticity to dynamically reconfigure the context manifold without weight updates. The approach enables path-dependent heterogeneous evolution while adhering to stringent governance constraints. We introduce the Structural Intelligence governance protocol, which defines auditable, reversible, and topologically continuous evolutionary trajectories, and provide falsifiable criteria alongside working examples to establish a theoretical foundation for heterogeneous intelligent ecosystems.

0 citationsRead paper

A deep learning approach for pricing convertible bonds with path-dependent reset and call provisions

May 12, 2026

This study addresses the pricing challenge of convertible bonds featuring path-dependent downward reset provisions and issuer call options by proposing a novel integration of path-dependent partial differential equations (PPDEs) with deep learning. The method employs a discrete-time dynamic programming framework implemented via neural networks to efficiently approximate conditional expectations, thereby handling complex path dependencies. It demonstrates robust performance across multiple underlying asset dynamics—including geometric Brownian motion, CEV, and Heston models—and accurately captures both prices and sensitivities in an empirical application to China CITIC Bank’s convertible bond. The analysis yields three key economic insights: contractual terms predominantly drive valuation; the call provision substantially depresses bond value; and, counterintuitively, the downward reset mechanism can lead to lower prices rather than enhancing investor protection.

0 citationsRead paper

Recent publications

Latest Papers

From Application-Layer Simulation to Native Meta-Architecture: Structural Tension as an Endogenous Driver for Heterogeneous AI Evolution

Jul 07, 2026

Current large language models lack intrinsic evolutionary capacity, relying instead on external prompt engineering that often leads to homogenization. This work proposes a native meta-architecture that, for the first time, leverages structural tension as an endogenous driving force, integrating offline recursive loops with inference-time plasticity to dynamically reconfigure the context manifold without weight updates. The approach enables path-dependent heterogeneous evolution while adhering to stringent governance constraints. We introduce the Structural Intelligence governance protocol, which defines auditable, reversible, and topologically continuous evolutionary trajectories, and provide falsifiable criteria alongside working examples to establish a theoretical foundation for heterogeneous intelligent ecosystems.

0 citationsRead paper

A deep learning approach for pricing convertible bonds with path-dependent reset and call provisions

May 12, 2026

This study addresses the pricing challenge of convertible bonds featuring path-dependent downward reset provisions and issuer call options by proposing a novel integration of path-dependent partial differential equations (PPDEs) with deep learning. The method employs a discrete-time dynamic programming framework implemented via neural networks to efficiently approximate conditional expectations, thereby handling complex path dependencies. It demonstrates robust performance across multiple underlying asset dynamics—including geometric Brownian motion, CEV, and Heston models—and accurately captures both prices and sensitivities in an empirical application to China CITIC Bank’s convertible bond. The analysis yields three key economic insights: contractual terms predominantly drive valuation; the call provision substantially depresses bond value; and, counterintuitively, the downward reset mechanism can lead to lower prices rather than enhancing investor protection.

0 citationsRead paper