In this hands-on session we’ll work with several datasets, such as MNIST handwritten digit dataset, Traffic Sign dataset, and build progressively more complex models to recognize images.
(1) We’ll use classifiers from Scikit-Learn;
(2) Implement LeNet-5 convolutional neural network in Keras/Tensorflow.
(3) Use transfer learning technique based on Inception or ResNet architectures in Keras/Tensforflow.
We will analyze and compare performance of each model and dataset combination, and discuss various issues related to data preparation and training.
This is a hands-on session, so bring your laptop, or use your instance at AWS for the duration of the session. I will publish a docker image with all libraries and datasets shortly, to simplify the setup.
BIO:
Alex Kalinin, VP Product and Data Science @ home.ai
https://www.linkedin.com/in/alexkalinin/