Convolutional Neural Networks

Explore convolutional neural networks in this comprehensive lecture from MIT's Introduction to Deep Learning course. Dive into deep computer vision, covering topics from basic image representation to advanced CNN architectures. Learn about manual and learned feature extraction, convolution operations, pooling, and the application of CNNs for classification tasks. Discover how CNNs are trained using backpropagation and their performance on the ImageNet dataset. Examine cutting-edge applications in semantic segmentation, image captioning, and real-world impacts in face detection, self-driving cars, and healthcare. Gain a solid foundation in deep learning techniques for computer vision through this in-depth presentation by lecturer Ava Soleimany.

Intro
Images are Numbers
Tasks in Computer Vision
High Level Feature Detection
Manual Feature Extraction
Learning Feature Representations
Fully Connected Neural Network
Using Spatial Structure
Applying Filters to Extract Features
Feature Extraction with Convolution
Filters to Detect X Features
The Convolution Operation
Producing Feature Maps
Convolutional Layers: Local Connectivity
Introducing Non-Linearity
Pooling
CNNs for Classification: Feature Learning
CNNs for Classification: Class Probabilities
CNNs: Training with Backpropagation
ImageNet Dataset
ImageNet Challenge: Classification Task
An Architecture for Many Applications
Beyond Classification
Semantic Segmentation: FCNs
Driving Scene Segmentation
Image Captioning using RNNS
Impact: Face Detection
Impact: Self-Driving Cars
Impact: Healthcare
Deep Learning for Computer Vision: Summary