Project Info:

Advances in using deep learning for computer vision tasks have been primarily driven by more complex neural network architectures. While this has resulted in material performance gains, it has also led to over-parameterized models that require significant computational resources. Deploying these complex models in resource constrained environments, such as mobile applications, has been challenging. In this research, we explore two techniques for model compression: pruning and quantization, which seek to balance model performance and computational costs. Our experiments prove that both techniques in isolation and a combination of the two are effective ways to significantly reduce model size and complexity without a corresponding degradation in performance on the CIFAR-10 dataset

Project Details:

• Technologies: Pytorch, Google Colab, Matplotlib
• Date: Jan. 2021 - May 2021
• Sources:Report Link