The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Applied AI Research at AWS with Alex Smola - #487
May 27, 2021
In this engaging discussion, Alex Smola, Vice President and Distinguished Scientist at AWS AI, explores cutting-edge AI research. He delves into deep learning on graphs, highlighting its role in enhancing data interpretation and applications like fraud detection. Alex also discusses the significance of AutoML, designed to make machine learning more accessible. He introduces Granger causality in causal modeling and shares insights about the growing AWS ML Summit, showcasing speaker highlights and exciting trends in AI.
AI Snips
Chapters
Transcript
Episode notes
PageRank's Random Surfer
- Alex Smola explains how PageRank, Google's foundational algorithm, was a stroke of genius.
- It mimics a random surfer's journey through web pages, determining importance by visit frequency.
Deep Learning on Graphs: Vertex Updates
- Deep learning on graphs learns vertex update functions, improving upon manually designed algorithms like PageRank.
- This allows for learning from data how to update a vertex's representation based on its neighbors.
Knowledge Graphs from Text
- Smola's team used cycle-consistent training to extract knowledge graphs from text, similar to image-to-image translation.
- This approach bypasses costly human annotation by ensuring consistency when translating between text and graph representations.