Daniel S Levine just posted an early draft of his brilliant book.

It's clear and readable.

Different systems have different qualities: speed, precision, discovery of novelty.

READ LEVINE.

Sam Leven

R u referring this book

Introduction to Neural and Cognitive Modeling 2nd Edition

by Daniel S. Levine (Author)

Yes. Among the many challenges facing neural network analysis is the trade-off between speed and texture; RL is relatively fast, SOM finds patterns, ART discovers. Levine's 3d edition provides depth and clarity, including new work.

Such an universal model matching all your requirements are simplicial complexes (topological graphs). Follow:

• J. Pfalzgraf, " Modeling Connectionist Networks: Categorical, Geometric Aspects (Towards “Homomorphic Learning”) ", 2004 - https://booksc.xyz/book/45352471/e19061
• Salnikov et al., " Simplicial complexes and complex systems ", 2018 -

  Preprint Simplicial complexes and complex systems

From simplicial complexes may be derived the neural network zoo and much more: http://www.asimovinstitute.org/neural-network-zoo/

Stéphane Breton

thank you so much for sharing the valuable info

Other source of information by following this related issue on RS: https://www.researchgate.net/post/Why_is_it_popular_to_use_machine_deep_learning_to_solve_PDEs

Hi Sandhya,

As a summary of my previous posts, an universal solution for implementing ANNs can be roughly formalized by the equation:

Universal ANNs = [possibly Stacked-] Combinatorial Maps + Multivector Algebra + Conformal (Projective) Geometry + [possibly Multiresolution] Solver (AMG/GMG, Lifting Scheme, Cell Method, Virtual Element Decomposition)

Most artificial neural networks have already been transposed to multivector algebra.

Thank you Stéphane Breton …