Continual Learning of Dense Signals

<a href="https://embed.stanford.edu/iframe/?url=https%3A%2F%2Fpurl.stanford.edu%2Frw194xv5100" class="su-underline">Show Content</a>

Abstract/Contents

Abstract: Coordinate networks are neural networks that parameterize low-dimensional signals like images and signed distance functions. They fail to sequentially learn on a stream of signals as they catastrophically forget the earlier part of the stream. This paper proposes a method to mitigate forgetting based on a hash encoding coordinate network. Specifically, the hash table is augmented with a cache data structure along with additional regularizations. Experiments show that this method eliminates forgetting when the network learns dense signals for different data collection patterns, approaching the empirical upper bound with little margin.

Type of resource	text
Publication date	July 20, 2023

Author	Dong, Zhengyang
Thesis advisor	Wetzstein, Gordon

DOI	https://doi.org/10.25740/rw194xv5100
Location	https://purl.stanford.edu/rw194xv5100

Use and reproduction: User agrees that, where applicable, content will not be used to identify or to otherwise infringe the privacy or confidentiality rights of individuals. Content distributed via the Stanford Digital Repository may be subject to additional license and use restrictions applied by the depositor.
License: This work is licensed under a Creative Commons Attribution Non Commercial 4.0 International license (CC BY-NC).

Preferred citation: Dong, Z. (2023). Continual Learning of Dense Signals. Stanford Digital Repository. Available at https://purl.stanford.edu/rw194xv5100. https://doi.org/10.25740/rw194xv5100.

Undergraduate Theses, School of Engineering

Loading usage metrics...