Realtime Hybrid Reinforcement Learning at Scale

Explore a cutting-edge approach to Next Best Action (NBA) in this conference talk from Strange Loop. Dive into the challenges of designing an AI-powered NBA engine that handles incomplete historical feedback, adapts to dynamic actions, optimizes for complex business objectives, and operates in real-time at massive scale. Learn how Salesforce Marketing Cloud Einstein addresses these issues through a hybrid model based on reinforcement learning, balancing online and offline learning. Discover the utilization of distributed big data processing technologies for large-scale training and prediction, and examine an offline evaluation mechanism providing bounded expected performance. Gain insights into the implementation of AI initiatives at scale, processing over 7 billion monthly unique users and making trillions of weekly predictions. Understand the applications of this technology in data management, real-time bidding, intelligent marketing, anti-fraud, and anti-money laundering efforts.

Introduction
Data
Artificial Intelligence
Salesforce
Marketing Cloud
Email Frequency
Personalization
Events
Interaction Studio
The Ghost
Context Remote Error Bandit
Existing Techniques
Challenges
Purpose
Use Case
Rolling Valuation