lattice-rnn-pruning

What problem does it solve?

This method reduces RNN size and computes pruning strategies by modeling neurons as a poset and constructing a dependency lattice to identify meet-irreducible neurons that should be preserved to maintain connectivity.

Core Features & Use Cases

• Poset modeling of neurons and dependency lattice construction for pruning guidance
• Identification of meet-irreducible neurons to preserve functional connectivity
• Support for multi-layer RNN pruning while maintaining sequence modeling performance
• Useful for model compression, edge deployment, and inference efficiency

Quick Start

Provide a trained PosetRNN and run prune with a target_sparsity to obtain the pruned matrix and kept neuron indices.