Courses

In this course, you'll leverage student engagement data to create a photo and text recommendation app similar to Instagram/Twitter. This app will utilize AI-generated photos and text and require you to recommend a feed from over 500,000 pieces of AI generated content. We'll explore various techniques to achieve this, including, but not limited to: Candidate Generation (Collaborative filtering, Trending, Cold start, N-tower neural network models, Cross-attention teachers, Distillation, Transfer learning, Random graph walking, Reverse indexes, LLMs as embedding), Filtering (Small online models, Caching, Deduplication, Policy), Prediction/Bidding (User logged activity based prediction (time-series), Multi-gate mixture of experts (MMOE), Regularization, Offline/Online evaluation (NDCG, p@k, r@k), Boosted Trees, Value Based Bidding), Ranking (Re-ranking, Ordering, Diversity, Enrich/Metadata/Personalization, Value Functions), Misc (Data Privacy and AI Ethics, Creator Based Models, Declared, Explicit and implicit topics, Explore/Exploit, Interpret/Understand/Context/Intention).
These concepts are applicable to various recommendation systems, from e-commerce to travel to social media to financial modeling. The instructor's experience at Uber Eats, Facebook, Instagram, and Google will provide valuable insights into real-world use cases.