Probabilistic Signature Inversion: Learning Conditional Distributions from Truncated Signatures
Researchers have developed a new probabilistic approach to invert truncated signatures, a method used to represent continuous-time paths. This technique reframes the ill-posed problem of recovering a path from its signature as learning a conditional distribution. The proposed method utilizes a signature-conditioned flow matching model and establishes a theoretical baseline for reconstruction error, which is then validated through experiments on financial data. AI
IMPACT Establishes a new probabilistic framework for signature inversion, potentially improving path reconstruction in time-series analysis.