Track trials, findings, and clinical evidence as it lands.
Search papers, labs, and topics across Lattice.
The authors investigate how weight location and local network context dictate reconstruction quality under pruning by applying controlled, pre-retraining structural edits across 120 convolutional U-Nets and Vision Transformers for MRI. While standard pruning benchmarks treat equal sparsity levels interchangeably, this work shows that computational role and weight compatibility critically govern performance independent of raw parameter counts. Crucially, preserving high-resolution paths and the initial convolution drives U-Net fidelity, whereas transformer pruning quality depends heavily on surrounding trained weights—causing identical masks to invert quality rankings when context shifts.
The exact same pruning mask can invert from best- to worst-performing depending purely on surrounding weight context, exposing how naive sparsity metrics fail to predict real model degradation in dense vision architectures.
Pruning reduces the number of weights in magnetic resonance imaging (MRI) reconstruction networks. Equal sparsity, however, can retain weights with different computational roles and different compatibility with the trained network. We study these effects across 120 convolutional U-Net and vision transformer models using controlled edits evaluated before retraining. In U-Net, preserving high-resolution computation improves reconstruction at equal deletion counts across all tested sparsity levels. Equal-operation controls reveal additional sensitivity of the first convolution, which operation count alone cannot explain. In transformers, retained pretrained weight energy correlates with quality within sparsity levels, yet the same masks reverse their reconstruction ranking when the surrounding trained weights change. At 90\% sparsity, matching intermediate output scales largely removes a high-energy-mask penalty while retaining an advantage for the network's original mask. Thus sparsity alone does not characterize reconstruction quality: architecture-specific descriptors are informative, but mask quality can still depend on the surrounding trained network.