Module 5- Part 3- What CNN architecture? Interpretable CNN and Transfer Learning in computer vision
Relevant playlists: Machine Learning Concepts, simply explained: • Machine Learning Concepts (shorter videos) Deep Learning Concepts, simply explained: • Deep Learning Codes and Concepts (Simply E... Instructor: Pedram Jahangiry All of the slides and notebooks used in this series are available on my GitHub page, so you can follow along and experiment with the code on your own. https://github.com/PJalgotrader Lecture Outline: 0:00 Roadmap and recap 1:28 Convnet architecture 4:05 MHR formula (Modularity, Hierarchy, Reuse) 6:10 Convnet architecture best practices 7:24 Residual connections 9:43 Batch Normalization 12:20 Depthwise separable convolutions 15:30 Best practices summary 18:47 Interpreting convnets (visualizing intermediate convnet output, filters and CAM: class activation map) 29:30 Classical CNN architectures 30:50 LeNet-5 33:28 AlexNet 34:49 VGG16 37:11 ResNet 39:51 Comparing the performance of classical CNN models on ImageNet data 43:00 State of the art models (as of March 2023) 44:00 Transfer learning (feature extraction, fine tuning)
