The cold start problem is the difficulty of making effective recommendations or decisions when a new action, item, user, or system has little or no interaction data.
A new ad creative or customer type has little or no interaction data, leaving a bandit unsure how to act. Without a fair chance to learn, new options can be buried by established ones. section 5.6
Give new options a chance with:
Optimistic priors that give new arms the benefit of the doubt. section 5.6
Warm starts that borrow knowledge from similar arms or users. section 5.6
How Applied Reinforcement Learning covers it
How deep it goes
Recommends optimistic priors and warm starts but does not detail their implementation.
Examples
Uses a new ad creative and a new customer type as cold-start cases.
What you can do after
Apply optimistic priors or warm starts to give new options a fair chance.
How it connects
Connects cold starts to bandits, exploration, and personalization.
AI-Powered SearchThe cold start problem is the difficulty of making reliable recommendations for new users or items that have little or no interaction data.
The Art of AI Product DevelopmentThe cold start problem is the challenge of making useful machine-learning or recommendation decisions when little or no relevant data is initially available.
Grokking Software ArchitectureA cold start is the startup latency incurred when a serverless runtime must initialize an execution environment and load function code before handling an invocation.
semantic matchingaddressed by · Semantic matching is the comparison of content by meaning to identify related items even when their wording differs.
content-based recommenderaddressed by · A content-based recommender suggests items whose features resemble those of items matching a user’s interests or prior interactions.
behavior-based recommendercontrasts with · A recommender that matches users and items according to overlapping interactions from similar users with similar items.
performance requirementused by · A performance requirement specifies a measurable expectation or constraint on a system’s speed, responsiveness, throughput, capacity, or resource use.
This explainer was generated by Manning's knowledge graph from the text of Applied Reinforcement Learning. The section links point to the parts of the book it draws on.