Big2Small: A Unifying Neural Network Framework for Model Compression
This work addresses the need for a principled approach to model compression for AI practitioners, offering a novel framework that unifies existing techniques, though it is incremental in building upon prior methods.
The paper tackles the fragmentation in model compression by proposing a unifying mathematical framework based on measure theory and introducing Big2Small, a data-free compression method that uses Implicit Neural Representations to encode weights, achieving competitive accuracy and compression ratios in image classification and segmentation tasks.
With the development of foundational models, model compression has become a critical requirement. Various model compression approaches have been proposed such as low-rank decomposition, pruning, quantization, ergodic dynamic systems, and knowledge distillation, which are based on different heuristics. To elevate the field from fragmentation to a principled discipline, we construct a unifying mathematical framework for model compression grounded in measure theory. We further demonstrate that each model compression technique is mathematically equivalent to a neural network subject to a regularization. Building upon this mathematical and structural equivalence, we propose an experimentally-verified data-free model compression framework, termed \textit{Big2Small}, which translates Implicit Neural Representations (INRs) from data domain to the domain of network parameters. \textit{Big2Small} trains compact INRs to encode the weights of larger models and reconstruct the weights during inference. To enhance reconstruction fidelity, we introduce Outlier-Aware Preprocessing to handle extreme weight values and a Frequency-Aware Loss function to preserve high-frequency details. Experiments on image classification and segmentation demonstrate that \textit{Big2Small} achieves competitive accuracy and compression ratios compared to state-of-the-art baselines.