Minimum Description Length Hopfield Networks. Abudy, M., Lan, N., Chemla, E., & Katzir, R. In Special Collection on Associative Memory and Hopfield Networks in PRX Life, 2023. To appear
Minimum Description Length Hopfield Networks [link]More  abstract   bibtex   
Associative memory architectures are designed for memorization but also offer, through their retrieval method, a form of generalization to unseen inputs: stored memories can be seen as prototypes from this point of view. Focusing on Modern Hopfield Networks (MHN), we show that a large memorization capacity undermines the generalization opportunity. We offer a solution to better optimize this tradeoff. It relies on Minimum Description Length (MDL) to determine during training which memories to store, as well as how many of them.
@inproceedings{MDLHN,
	abstract = {Associative memory architectures are designed for memorization but also offer, through their retrieval method, a form of generalization to unseen inputs: stored memories can be seen as prototypes from this point of view. Focusing on Modern Hopfield Networks (MHN), we show that a large memorization capacity undermines the generalization opportunity. We offer a solution to better optimize this tradeoff. It relies on Minimum Description Length (MDL) to determine during training which memories to store, as well as how many of them.},
	author = {Matan Abudy and Nur Lan and Emmanuel Chemla and Roni Katzir},
	booktitle = {Special Collection on Associative Memory and Hopfield Networks in PRX Life},
	date-added = {2023-12-20 15:46:34 +0100},
	date-modified = {2023-12-20 15:48:43 +0100},
	note = {To appear},
	title = {Minimum Description Length Hopfield Networks},
	url_more = {https://arxiv.org/abs/2311.06518},
	year = {2023}}

Downloads: 0