Classifications in modular restricted Boltzmann machines
What it is
This paper analyzes a modular neural network composed of coupled Hopfield models and restricted Boltzmann machines (RBMs). It demonstrates that competitive coupling between modules enables pattern disentanglement and proves that specific weight initializations act as fixed points for classification tasks involving orthogonal patterns.
The problem it solves
Standard neural networks often struggle to disentangle mixed or superimposed patterns, and the theoretical stability of training dynamics for modular competitive architectures is not well understood.
What you could use it for
- Joint classification and disentanglement of mixed input patterns
- Designing competitive associative memory networks
- Theoretical analysis of RBM training dynamics
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.