ICLR 2019 Expo Panel
AI Research using PyTorch: Bayesian Optimization, Billion Edge Graphs and Private Deep Learning
We'll walk through how to code the latest research in PyTorch as well as dive into frameworks, built on top of PyTorch, focused on large scale graph embeddings and Bayesian optimization. We'll finish by discussing the latest in Privacy and AI and get hands on with a library that enables private deep learning on PyTorch. Big Graph PyTorch Big Graph can be used for generating embeddings from large-scale graph-structured data. Users can generate embeddings of entities that can be described as a set of nodes connected via a set of edges where the edges may represent different relationships. For example, users in a social graph are connected to each other via different types of relationships (friends, went to the same university). The relationships among the users can be captured in their embeddings in such a way that the distance between a pair of users in the original graph is generally preserved in the distance based on their embedding vectors.
Bayesian Optimization Botorch (“Bayesian Optimization in PyTorch”) is a library for Bayesian Optimization. Bayesian Optimization is an established technique for sequential optimization of costly-to-evaluate black-box functions. It can be applied to a wide variety of problems, including Machine Learning (tuning algorithms' hyper-parameters), robotics, and A/B testing. Bayesian Optimization is routinely used for sequential A/B testing, optimizing backend services, and improving performance of Machine Learning models.