Dynamic Co-Optimization Compiler: Leveraging Multi-Agent Reinforcement Learning for Enhanced DNN Accelerator Performance
This addresses the challenge of optimizing DNN deployments for varied hardware, offering significant performance improvements for ML practitioners, though it is incremental as it builds on existing co-optimization and reinforcement learning approaches.
This paper tackles the problem of efficiently mapping deep neural networks onto diverse hardware platforms by introducing a Dynamic Co-Optimization Compiler (DCOC) that uses multi-agent reinforcement learning, achieving up to 37.95% higher throughput and up to 42.2% faster optimization time compared to existing methods.
This paper introduces a novel Dynamic Co-Optimization Compiler (DCOC), which employs an adaptive Multi-Agent Reinforcement Learning (MARL) framework to enhance the efficiency of mapping machine learning (ML) models, particularly Deep Neural Networks (DNNs), onto diverse hardware platforms. DCOC incorporates three specialized actor-critic agents within MARL, each dedicated to different optimization facets: one for hardware and two for software. This cooperative strategy results in an integrated hardware/software co-optimization approach, improving the precision and speed of DNN deployments. By focusing on high-confidence configurations, DCOC effectively reduces the search space, achieving remarkable performance over existing methods. Our results demonstrate that DCOC enhances throughput by up to 37.95% while reducing optimization time by up to 42.2% across various DNN models, outperforming current state-of-the-art frameworks.