Javascript must be enabled for the correct page display

Long Short-Term Memory Recurrent Neural Networks and a Feedforward alternative for the Classification of Sperm Whale-emitted Clicks

Zonca, Noam (2021) Long Short-Term Memory Recurrent Neural Networks and a Feedforward alternative for the Classification of Sperm Whale-emitted Clicks. Bachelor's Thesis, Artificial Intelligence.

[img]
Preview
Text
BachelorThesis_NoamZonca_s3482065.pdf

Download (2MB) | Preview
[img] Text
toestemming.pdf
Restricted to Registered users only

Download (131kB)

Abstract

In the study of cognition, humans make useful subjects for inspection. However, recent findings show that Sperm Whales possess surprisingly advanced and comparable cognitive abilities. Unfortunately, due to their unfriendly natural environment, these mammals are hard to study and understand. Nonetheless, using data-intensive bioacoustic methods and machine learning, the hope of understanding their complex behaviour through their language is stronger than ever. Gruber, Bronstein, Wood, Gero, and Bermant (2019) applied Recurrent Neural Networks to sperm whale emitted codas and showed an astonishing above 90% accuracy at recognising coda type, vocal clan and even individual specimen ID. Upon confirming their study by reproducing the machine learning process, a Feedforward Sliding Window architecture has also been applied to experiment its effectiveness in the classification and recognition of sperm whale codas. Moreover, by re-implementing, re-tuning and re-training the models by Gruber et al. (2019), as well as implementing, tuning and training the feedforward sliding window approach, a comparison of these approaches with the Gruber et al. (2019) results is carried out. Results found show that the Gruber et al. (2019) study can be confirmed and that the recurrent approach is hard to beat, but the feedforward sliding window approach can also shine, with accuracies around the 84% mark.

Item Type: Thesis (Bachelor's Thesis)
Supervisor name: Wolf, B.J. and Netten, S.M. van
Degree programme: Artificial Intelligence
Thesis type: Bachelor's Thesis
Language: English
Date Deposited: 16 Jul 2021 12:22
Last Modified: 16 Jul 2021 12:22
URI: https://fse.studenttheses.ub.rug.nl/id/eprint/25283

Actions (login required)

View Item View Item