Intro to ConvNets
summary
A practical introduction to convolutional nets that gives concrete claims and commands not platitudes: explains forward/backprop and regularisation, shows training on 60,000 32x32 color images in 10 classes stored as Python pickle batches, demonstrates the cuda-convnet Python bindings on NVIDIA GPUs, and walks the layer definitions, pooling, dropout and dropconnect settings, plus show_net/show_filters utilities. Reports intermediate results (about 36.7% classification error at an early epoch), describes checkpointing, batch setup and multi‑GPU/data/model parallelism tradeoffs.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:11Classification of labeled data
- 3:50Problem of object recognition in images
- 6:42Main reason for the resurgence
- 7:47Higher level overview of a neural network
- 10:20How does this work?
- 14:20Multi-class logistic classification
- 16:02Backward propagation
- 17:15Training a neural network
- 18:15Feature extraction in images
- 20:10What is a convolution?
- 23:10Convolutional neural networks (in Python)
- 24:42Cuda-convnet
- 35:05Regularisation
- 36:58Parallelising CNNs
- 40:07End of talk
- 40:40Questions
related talks
51mA Practical Introduction to Convnets for Image Classification
27mDemystifying Deep Neural Networks
54mIntroduction to Understanding Deep Learning
28mDemystifying Deep Neural Nets
55mA Tour of Deep Learning With C++
30mMachine Learning & Neural Networks Demystified
01h 27mHands-On Computer Vision with PyTorch
24mConvolutional Neural Networks in Action