Cognitive sampling: The power of harnessing sampling strategies with machine learning techniques for enhanced machine classification and machine interpretation of acoustic datasets

Abstract

Listen

Translate article

Machine learning has received increased popularity in the acoustical community for its ability to interpret, classify and predict information autonomously from large-scale datasets. Beyond traditional supervised and semi-supervised learning architectures such as deep learning, and ensemble approaches, unsupervised learning networks such as autoencoders are being used to discover efficient data codings that enable hitherto unforeseen feature spaces. Unrelated yet deeply significant to the success of these myriad approaches is the spectral resolution of the signal itself as it has implications on the richness of information used by the learning architecture. Furthermore, much remains to be done in interpreting machine-learnt feature representations by domain experts and linking the domain knowledge to the machine-learnt knowledge. This talk will introduce a new way of enabling this link, by harnessing the power of compressive sampling and more generally, non-uniform sampling strategies and information theory to popular machine learning techniques. Specific applications will include a variety of acoustical applications involving spectral feature generation and interpretation.