Logistic Disjunctive Normal Network (LDNN)

This is an implementation of Logistic Disjunctive Neural Networks as proposed in the paper listed under References.

Implementation State

Basic Functionality

• data structures
• read and write files
• initialize
• evaluate/classify
• gradient decendent learning
• (optional) convolutional network

Optimization

• optimize memory
• parallelize initialization
• parallelize gradient decent learning
• parallelize classification (only preferable on very big networks)

How to use

run make to build.
run make test to run a simple test.
test.c shows how to create, initialize, train und evaluate a LDNN.

References

M. Sajjadi, M. Seyedhosseini, T. Tasdizen "Disjunctive Normal Networks" arXiv:1412.8534 [cs.LG] (2014) http://arxiv.org/abs/1412.8534