Cart 0
EQUIVARIANT & COORDINATE INDEPENDENT CONVOLUTIONAL NETWORKS
Click to zoom

Share this book

EQUIVARIANT & COORDINATE INDEPENDENT CONVOLUTIONAL NETWORKS

Book Details

Format Hardback or Cased Book
ISBN-10 9819806623
ISBN-13 9789819806621
Publisher World Scientific Publishing Co Pte Ltd
Imprint World Scientific Publishing Co Pte Ltd
Country of Manufacture GB
Country of Publication GB
Publication Date Feb 9th, 2026
Print length 592 Pages
Ksh 29,700.00
Werezi Extended Catalogue 0 in stock

Delivery Location

Delivery fee: Select location

Secure
Quality
Fast

What is the appropriate geometric structure for neural networks that process spatial signals on Euclidean spaces or more general manifolds? This question takes us on a journey which leads to a gauge field theory of convolutional networks.

Feature vector fields: The spatial signals we are interested in are fields of feature vectors. Feature fields allow to describe data like images, audio, videos, point clouds, or tensor fields, such as fluid flows and electromagnetic fields.

Equivariant networks commute with actions of some symmetry group on their feature spaces. The relevant group actions in this work are geometric transformations of feature fields, like translations, rotations, or reflections of images. Equivariant models generalize everything they learn over the considered group of transformations. This property makes them significantly more data efficient, interpretable, and robust in comparison to non-equivariant models.

Convolutional Neural Networks (CNNs) are the most common network architecture for processing feature fields. Conventional CNNs operate on Euclidean spaces and are translation equivariant, i.e. position independent. This work explains how to extend CNNs to be equivariant under more general symmetries of space.

Coordinate independence: Manifolds are in general not equipped with a canonical choice of coordinates. Feature fields and neural network layers are hence required to be coordinate independent, that is, expressible relative to different frames of reference. The ambiguity of local frames represents the gauge freedom of our neural field theory. We show that the demand for coordinate independence requires CNNs to be equivariant under local gauge transformations.


Get EQUIVARIANT & COORDINATE INDEPENDENT CONVOLUTIONAL NETWORKS by at the best price and quality guaranteed only at Werezi Africa's largest book ecommerce store. The book was published by World Scientific Publishing Co Pte Ltd and it has pages.

Mind, Body, & Spirit

Shopping Cart

Africa largest book store

Sub Total:
Ebooks

Digital Library
Coming Soon

Our digital collection is currently being curated to ensure the best possible reading experience on Werezi. We'll be launching our Ebooks platform shortly.