# Welcome!
References
Motivation
Bottleneck widens: Memory and compute requirements increase
Sparse NNs
Quantized NNs
Boolean circuit
Zoom out again
Idea: LGNs
Which gate?
Which connections?
Learning gates
Problem: Discrete gate choices
Probabilistic logics
Rounding
Hard Rounding
Learning connections
Classification
Differentiable Training
Gradient Analysis
Reconsider how resinit stabilizes the training curve
Only natural vanishing gradient?
Go back to our grad analysis
Inappropriate Regularization
Group sum