Why Real-Time Machine Learning Thrives on Fresh Data

Explore the importance of fresh data in real-time machine learning applications through this 34-minute presentation by Chip Huyen, co-founder of Claypot AI. Delve into the value of up-to-date information, various architectures for online prediction, and the challenges faced in implementing these systems. Learn about use cases for real-time ML, architectures optimized for online predictions considering feature computation, prediction, and request response times. Discover strategies to overcome key challenges in online prediction systems, including balancing latency with feature freshness, maintaining accuracy, and managing streaming infrastructure. Gain insights from Huyen's extensive experience in the field, including her work at Snorkel AI and NVIDIA, as well as her teachings at Stanford University on Machine Learning Systems Design.

Introduction
Legacy and Stillness
Latency
Predict Inconsistency
Challenges
Backfilling
Unified Instrument