Dualistic geometry of the manifold of higher-order neurons
β Scribed by Shun-ichi Amari
- Publisher
- Elsevier Science
- Year
- 1991
- Tongue
- English
- Weight
- 940 KB
- Volume
- 4
- Category
- Article
- ISSN
- 0893-6080
No coin nor oath required. For personal study only.
β¦ Synopsis
A
.w of nerlral networks. in particular rite set of all rhe nellrotzs of higher-order. f~mtl.s a geortlerrical manifold. Specifically, the .set.s N,(k = 1, 2.
.) of the krh order t1euron.s constitute a hierarchy of manifolds N,, > N,, , > '.' > N, , where n is rhe number of'itlputs. A natural geometry is inrroduceri to NL and characteristics of higher-order neurons are studied therefrom. A Rietnannian metric is defined itI N, and a dual pair of a,ffine contlecrions are introduced in these manifolds. A higher-order neuron realizes a transfortnatiotl from vector itlpuCs to a scalar ourpur. Given a rramfortnation.
π SIMILAR VOLUMES
It has been clarified recently that an r-th order Lagrangian on a fibred manifold Y X does not determine a unique PoincarΓ©-Carran fonn provided dimX> 1 and r>2, [1], [4], [6], [9], [10]. To make this fact more transparent, we introduced a new operation generalizing the formal exterior differentiatio
A two-dimensional finite element technique, corroborated with experimental results, was used to explore the acoustic characteristics of 90 ~ bends when cross modes propagate. The influence of changes in geometry on these characteristics was established over a range of inner and outer radii and the e