Compression Strategies and Space Conscious Representations for Deep Neural Networks


Recent advances in deep learning have made available large, powerful convolutional neural networks (CNN) with state-of-the-art performance in several real-world applications. Unfortunately, these large-sized models have millions of parameters, thus they are not deployable on resource-limited platforms (e.g., where RAM is limited). Compression of CNNs becomes therefore a critical problem to achieve memory-efficient and possibly computationally faster model representations.  In this paper, we investigate the impact of lossy compression of CNNs by weight pruning and quantization, and lossless weight matrix representations based on source coding. We  tested several combinations of these techniques on four benchmark datasets for classification and regression problems, achieving compression rates up to 165 times, while preserving or improving the model performance."

Proceedings of the 25th International Conference on Pattern Recognition (ICPR)
Marco Frasca
Assistant professor

Researcher in Machine Learning and AI of the UNIMI unit

Dario Malchiodi
Associate professor

Professor of Data analytics and member of the unimi team