CoNNect: Connectivity-Based Regularization for Structural Pruning
This addresses the problem of maintaining connectivity during pruning for ML practitioners, though it appears incremental as it builds on existing structural pruning methods.
The paper tackles the problem of neural network pruning by introducing CoNNect, a differentiable regularizer that ensures connectivity between input and output layers while approximating L0 regularization. The result shows CoNNect improves classical pruning strategies and enhances state-of-the-art one-shot pruners like DepGraph and LLM-pruner.
Pruning encompasses a range of techniques aimed at increasing the sparsity of neural networks (NNs). These techniques can generally be framed as minimizing a loss function subject to an $L_0$ norm constraint. This paper introduces CoNNect, a novel differentiable regularizer for sparse NN training that ensures connectivity between input and output layers. We prove that CoNNect approximates $L_0$ regularization, guaranteeing maximally connected network structures while avoiding issues like layer collapse. Moreover, CoNNect is easily integrated with established structural pruning strategies. Numerical experiments demonstrate that CoNNect can improve classical pruning strategies and enhance state-of-the-art one-shot pruners, such as DepGraph and LLM-pruner.