Continuous Input Embedding Size Search For Recommender Systems: A Novel RL-Based Method for Memory-Efficient Recommendation
Core Concepts
Latent factor models in recommender systems can benefit from continuous input embedding size search to optimize memory efficiency and recommendation performance.
Continuous Input Embedding Size Search For Recommender Systems
Stats
10 million items into 256-dimensional vectors can exceed 9 GB memory consumption in a double-precision float system.
CIESS outperforms other methods under different memory budgets when paired with three popular recommendation models.
Quotes
"Latent factor models are the most popular backbones for today’s recommender systems owing to their prominent performance."
"Existing RL-based methods are restricted to highly discrete, predefined embedding size choices."
"CIESS is a versatile embedding size search approach that does not hold any assumptions on the backbone recommendation model."