Unveiling Invariances via Neural Network Pruning
This addresses the need for automated invariance learning in neural networks, offering a domain-specific improvement over handcrafted designs.
The paper tackles the problem of manually designing neural networks to handle invariances by proposing a pruning-based framework that learns data-dependent invariances, resulting in architectures that outperform dense networks in efficiency and effectiveness across vision and tabular datasets.
Invariance describes transformations that do not alter data's underlying semantics. Neural networks that preserve natural invariance capture good inductive biases and achieve superior performance. Hence, modern networks are handcrafted to handle well-known invariances (ex. translations). We propose a framework to learn novel network architectures that capture data-dependent invariances via pruning. Our learned architectures consistently outperform dense neural networks on both vision and tabular datasets in both efficiency and effectiveness. We demonstrate our framework on multiple deep learning models across 3 vision and 40 tabular datasets.