This document summarizes Fedor Borisyuk's presentation on image search at Facebook. It discusses: 1. An overview of Facebook's photo search product and the billions of images uploaded daily that it searches through. 2. The infrastructure and machine learning techniques used to perform large-scale image classification, ranking, and modeling similarity between queries and images. This includes convolutional neural networks, optical character recognition, and embedding models. 3. A deep dive into Facebook's approach to large-scale image classification, which uses ResNeXt models trained on billions of public images and hashtags to recognize objects and scenes. Noise handling techniques like label merging and sampling are also discussed. 4. Details on Facebook's