Fushing Hsieh, PhD, UC Davis
Title: From Learning Gait Signatures to Revelation of Multi-Scale Dependency in Musculoskeletal System
Summary: If human's seemingly born ability of recognizing individuals' gait signatures is indeed acquired by learning, how our brains could possibly achieve such complex tasks? Simple enough algorithmic computing is likely the answer. In this paper, we link gait's dynamic signatures to intrinsic structural dependency of multiple gait time series derived from our musculoskeletal system. Three kinds of manifestations of such dependency with distinct merits are explained and discussed. The first one based on tempo-free mutual conditional entropy and the second one based on Principle System-State Analysis (PSSA) are good for identification among many subjects. It is the third one based on Local-first then Global-second (L1G2) Coding Algorithm that can bring out complicate rhythmic patterns for idiosyncratic gait authentication. The revelation of two-layer coding scheme announces the duality between system dynamics and structural dependency along global system-state trajectory built by coupling several mechanism-specific local system-state trajectories. A ``passtensor'' is constructed to collectively represent all rhythmic-cycle-specific transition matrices among global system-states. The deterministic as well as stochastic structures embedded within such a passtensor make it the individual's intrinsic biometric traits. Through the L1G2's computing simplicity, we echo how our brain performs seemingly ordinary gait recognition. The implications of this revelation in sciences potentially go far beyond gait analysis.