Liquid Spiking Networks with Adaptive Structural Dynamism

Gavi Matam, Harshith Sai (2023) Liquid Spiking Networks with Adaptive Structural Dynamism. Master's Thesis / Essay, Artificial Intelligence.

Preview	Text Master_thesis_harshith_gavi.pdf Download (2MB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (127kB)

Abstract

Spiking Neural Networks (SNNs) represent an advanced class of neural networks, characterised by their ability to more closely mimic the temporal dynamics of biological neural systems. These networks were integrated with Liquid Time-Constant Networks to enhance the model's capability to retain information over extended time horizons, thereby performing well on time series tasks. In this thesis, this Liquid Spiking Network model was evaluated on an event-based audio classification task to establish an initial benchmark. A series of experiments were designed and conducted to assess the model and its hyperparameters, leading to the conclusion that the model achieves a test accuracy of 84.7%. Notably, when efficiency is considered, the model's global ranking improves, underscoring its real-world applicability. Post-evaluation, neurons were provided with the ability to dynamically choose their connections to examine the impact on performance and efficiency. Our findings suggest that this capability leads to a network compression of 10.5%. Although this compression marginally reduces accuracy by 3%, it offers a balanced trade-off between performance and efficiency compared to other state-of-the-art models.

Item Type:	Thesis (Master's Thesis / Essay)
Supervisor name:	Jaeger, H. and Abreu, S.
Degree programme:	Artificial Intelligence
Thesis type:	Master's Thesis / Essay
Language:	English
Date Deposited:	29 Nov 2023 10:33
Last Modified:	29 Nov 2023 10:33
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/31681

Actions (login required)

View Item