2013/12/16 by A. E. Allahverdyan, Allahverdyan, Armen E., Aram Galstyan +1
Computer Science · #FOS: Computer and information sciences #Gaussian Processes and Bayesian Inference #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Machine Learning and Algorithms #Neural Networks and Applications #Speech Recognition and Synthesis #Target Tracking and Data Fusion in Sensor Networks
paper · pdf · doi:10.48550/arxiv.1312.4551
openalex publication_date 2013/12/16 · openalex created_date 2025/10/24 · openalex updated_date 2026/07/28
We present an asymptotic analysis of Viterbi Training (VT) and contrast it\nwith a more conventional Maximum Likelihood (ML) approach to parameter\nestimation in Hidden Markov Models. While ML estimator works by (locally)\nmaximizing the likelihood of the observed data, VT seeks to maximize the\nprobability of the most likely hidden state sequence. We develop an analytical\nframework based on a generating function formalism and illustrate it on an\nexactly solvable model of HMM with one unambiguous symbol. For this particular\nmodel the ML objective function is continuously degenerate. VT objective, in\ncontrast, is shown to have only finite degeneracy. Furthermore, VT converges\nfaster and results in sparser (simpler) models, thus realizing an automatic\nOccam's razor for HMM learning. For more general scenario VT can be worse\ncompared to ML but still capable of correctly recovering most of the\nparameters.\n