Introduction to Convolutional Neural Networks for Image Recognition

Wednesday, October 3, 2018 - 17:30
TH 331
Alex Kalinin
In this presentation we will focus on the design of convolutional neural networks (CNNs), and how the relate to traditional, fully-connected neural network. We will look at the history of the research that lead to key elements of convolutional neural networks, such as convolutions and pooling. We will review the design of one of the first production-quality convolutional neural networks used for handwritten digit recognition — LeNet-5.

Alex ( leads AI/Machine Learning team at Sizmek - the largest independent buy-side advertising platform. The team develops cutting edge conversion models, recommender systems, models for automatic A/B testing, and many others. These models are applied in real time at scale, with billions of requests processed per
day. Previously, Alex worked in both startups and large companies. While at he led the team to develop home automation algorithms leveraging computer vision and convolutional neural networks. At Yahoo he led the development of the large-scale user acquisitions and analytics system supporting rapid growth of Yahoo Games business. Alex holds MS in Physics degree, and published several papers on image recognition and pattern detection.