Simplified Transition Matrix Estimation with Implicit Regularization for Enhanced Model Efficiency

We propose a simplified method that replaces the estimation of individual transition matrices for each sample by estimating a single overall transition matrix, in conjunction with implicit regularization. This approach significantly reduces computational complexity and enhances model efficiency without sacrificing performance. By leveraging the power of implicit regularization, our method effectively learns a shared representation across all samples, capturing the underlying transition dynamics in a more efficient manner.