2020/01/16 by Johan Pauwels, Pauwels, Johan, György Fazekas +3
Computer Science · Neuroscience · #Artificial Intelligence (cs.AI) #Audio and Speech Processing (eess.AS) #FOS: Computer and information sciences #FOS: Electrical engineering #Information Retrieval (cs.IR) #Machine Learning (cs.LG) #Music Technology and Sound Studies #Music and Audio Processing #Neural Networks and Applications #Neuroscience and Music Perception #Sound (cs.SD) #Time Series Analysis and Forecasting #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2001.06086
openalex publication_date 2020/01/16 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
In recent years, Markov logic networks (MLNs) have been proposed as a\npotentially useful paradigm for music signal analysis. Because all hidden\nMarkov models can be reformulated as MLNs, the latter can provide an\nall-encompassing framework that reuses and extends previous work in the field.\nHowever, just because it is theoretically possible to reformulate previous work\nas MLNs, does not mean that it is advantageous. In this paper, we analyse some\nproposed examples of MLNs for musical analysis and consider their practical\ndisadvantages when compared to formulating the same musical dependence\nrelationships as (dynamic) Bayesian networks. We argue that a number of\npractical hurdles such as the lack of support for sequences and for arbitrary\ncontinuous probability distributions make MLNs less than ideal for the proposed\nmusical applications, both in terms of easy of formulation and computational\nrequirements due to their required inference algorithms. These conclusions are\nnot specific to music, but apply to other fields as well, especially when\nsequential data with continuous observations is involved. Finally, we show that\nthe ideas underlying the proposed examples can be expressed perfectly well in\nthe more commonly used framework of (dynamic) Bayesian networks.\n