MaRCA: Multi-Agent Reinforcement Learning for Dynamic Computation Allocation in Large-Scale Recommender Systems
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866918266237091840 |
|---|---|
| author | Jiang, Wan Zang, Xinyi Zhao, Yudong Zou, Yusi Lu, Yunfei Tong, Junbo Liu, Yang Li, Ming Shi, Jiani Yang, Xin |
| author_facet | Jiang, Wan Zang, Xinyi Zhao, Yudong Zou, Yusi Lu, Yunfei Tong, Junbo Liu, Yang Li, Ming Shi, Jiani Yang, Xin |
| contents | Modern recommender systems face significant computational challenges due to growing model complexity and traffic scale, making efficient computation allocation critical for maximizing business revenue. Existing approaches typically simplify multi-stage computation resource allocation, neglecting inter-stage dependencies, thus limiting global optimality. In this paper, we propose MaRCA, a multi-agent reinforcement learning framework for end-to-end computation resource allocation in large-scale recommender systems. MaRCA models the stages of a recommender system as cooperative agents, using Centralized Training with Decentralized Execution (CTDE) to optimize revenue under computation resource constraints. We introduce an AutoBucket TestBench for accurate computation cost estimation, and a Model Predictive Control (MPC)-based Revenue-Cost Balancer to proactively forecast traffic loads and adjust the revenue-cost trade-off accordingly. Since its end-to-end deployment in the advertising pipeline of a leading global e-commerce platform in November 2024, MaRCA has consistently handled hundreds of billions of ad requests per day and has delivered a 16.67% revenue uplift using existing computation resources. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_24325 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | MaRCA: Multi-Agent Reinforcement Learning for Dynamic Computation Allocation in Large-Scale Recommender Systems Jiang, Wan Zang, Xinyi Zhao, Yudong Zou, Yusi Lu, Yunfei Tong, Junbo Liu, Yang Li, Ming Shi, Jiani Yang, Xin Information Retrieval Machine Learning Multiagent Systems H.3.3; I.2.11 Modern recommender systems face significant computational challenges due to growing model complexity and traffic scale, making efficient computation allocation critical for maximizing business revenue. Existing approaches typically simplify multi-stage computation resource allocation, neglecting inter-stage dependencies, thus limiting global optimality. In this paper, we propose MaRCA, a multi-agent reinforcement learning framework for end-to-end computation resource allocation in large-scale recommender systems. MaRCA models the stages of a recommender system as cooperative agents, using Centralized Training with Decentralized Execution (CTDE) to optimize revenue under computation resource constraints. We introduce an AutoBucket TestBench for accurate computation cost estimation, and a Model Predictive Control (MPC)-based Revenue-Cost Balancer to proactively forecast traffic loads and adjust the revenue-cost trade-off accordingly. Since its end-to-end deployment in the advertising pipeline of a leading global e-commerce platform in November 2024, MaRCA has consistently handled hundreds of billions of ad requests per day and has delivered a 16.67% revenue uplift using existing computation resources. |
| title | MaRCA: Multi-Agent Reinforcement Learning for Dynamic Computation Allocation in Large-Scale Recommender Systems |
| topic | Information Retrieval Machine Learning Multiagent Systems H.3.3; I.2.11 |
| url | https://arxiv.org/abs/2512.24325 |