Kaleidoscope: An Efficient, Learnable Representation For All Structured Linear Maps

Modern neural network architectures use structured linear transformations, such as low-rank matrices, sparse matrices, permutations, and the Fourier transform, to improve inference speed and reduce memory usage compared to general linear maps. However, choosing which of the myriad structured transfo...

Ausführliche Beschreibung

Gespeichert in:

Bibliographische Detailangaben
Veröffentlicht in:	arXiv.org 2021-01
Hauptverfasser:	Dao, Tri, Sohoni, Nimit S, Gu, Albert, Eichhorn, Matthew, Blonder, Amit, Leszczynski, Megan, Atri Rudra, Ré, Christopher
Format:	Artikel
Sprache:	eng
Schlagworte:	Accuracy Computer architecture Filter banks Fourier transforms Image classification Inference Language translation Linear transformations Matrix methods Model accuracy Neural networks Parameterization Permutations Pipelines Representations Sparse matrices Speech recognition Structured matrices
Online-Zugang:	Volltext
Tags:	Tag hinzufügen Keine Tags, Fügen Sie den ersten Tag hinzu!

Schreiben Sie den ersten Kommentar!