In this paper the theory and design of codes capable of correcting t insertion/deletion of the symbol 0 in each and every bucket of zeros (i. e., zeros in between two consecutive ones) are studied. It is shown that this problem is related to the zero error capacity achieving codes in limited magnitude error channel. Close to optimal non-systematic code designs and the encoding/decoding algorithms are described.
more »
« less
Zero-TPrune: Zero-Shot Token Pruning Through Leveraging of the Attention Graph in Pre-Trained Transformers
- Award ID(s):
- 2203399
- PAR ID:
- 10569650
- Publisher / Repository:
- IEEE
- Date Published:
- ISBN:
- 979-8-3503-5300-6
- Page Range / eLocation ID:
- 16070 to 16079
- Format(s):
- Medium: X
- Location:
- Seattle, WA, USA
- Sponsoring Org:
- National Science Foundation
An official website of the United States government

