Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Explainable Fuzzy Neural Network with Multi-Fidelity Reinforcement Learning for Micro-Architecture Design Space Exploration

Hanwei Fan, Ya Wang, Sicheng Li, Tingyuan Liang, Wei Zhang

TL;DR

Addresses interpretability and efficiency in micro-architecture design space exploration by combining an explainable FNN with a two-stage multi-fidelity reinforcement learning framework. An FNN learns interpretable design rules and can be steered by designers, while LF uses an analytical CPI model to rapidly explore and HF refines promising designs through RTL-based feedback. Empirical results on application-specific and general-purpose DSE show strong performance with limited high-fidelity budget and provide tangible rule-based insights. The framework is open-source, enabling reuse and further research in explainable architecture optimization.

Abstract

With the continuous advancement of processors, modern micro-architecture designs have become increasingly complex. The vast design space presents significant challenges for human designers, making design space exploration (DSE) algorithms a significant tool for $μ$-arch design. In recent years, efforts have been made in the development of DSE algorithms, and promising results have been achieved. However, the existing DSE algorithms, e.g., Bayesian Optimization and ensemble learning, suffer from poor interpretability, hindering designers' understanding of the decision-making process. To address this limitation, we propose utilizing Fuzzy Neural Networks to induce and summarize knowledge and insights from the DSE process, enhancing interpretability and controllability. Furthermore, to improve efficiency, we introduce a multi-fidelity reinforcement learning approach, which primarily conducts exploration using cheap but less precise data, thereby substantially diminishing the reliance on costly data. Experimental results show that our method achieves excellent results with a very limited sample budget and successfully surpasses the current state-of-the-art. Our DSE framework is open-sourced and available at https://github.com/fanhanwei/FNN\_MFRL\_ArchDSE/\ .

Explainable Fuzzy Neural Network with Multi-Fidelity Reinforcement Learning for Micro-Architecture Design Space Exploration

TL;DR

Addresses interpretability and efficiency in micro-architecture design space exploration by combining an explainable FNN with a two-stage multi-fidelity reinforcement learning framework. An FNN learns interpretable design rules and can be steered by designers, while LF uses an analytical CPI model to rapidly explore and HF refines promising designs through RTL-based feedback. Empirical results on application-specific and general-purpose DSE show strong performance with limited high-fidelity budget and provide tangible rule-based insights. The framework is open-source, enabling reuse and further research in explainable architecture optimization.

Abstract

With the continuous advancement of processors, modern micro-architecture designs have become increasingly complex. The vast design space presents significant challenges for human designers, making design space exploration (DSE) algorithms a significant tool for -arch design. In recent years, efforts have been made in the development of DSE algorithms, and promising results have been achieved. However, the existing DSE algorithms, e.g., Bayesian Optimization and ensemble learning, suffer from poor interpretability, hindering designers' understanding of the decision-making process. To address this limitation, we propose utilizing Fuzzy Neural Networks to induce and summarize knowledge and insights from the DSE process, enhancing interpretability and controllability. Furthermore, to improve efficiency, we introduce a multi-fidelity reinforcement learning approach, which primarily conducts exploration using cheap but less precise data, thereby substantially diminishing the reliance on costly data. Experimental results show that our method achieves excellent results with a very limited sample budget and successfully surpasses the current state-of-the-art. Our DSE framework is open-sourced and available at https://github.com/fanhanwei/FNN\_MFRL\_ArchDSE/\ .

Paper Structure

This paper contains 13 sections, 7 equations, 7 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Fuzzy Neural Networks for Micro-Architecture DSE
  3. Fuzzy Logic
  4. Fuzzy Neural Networks
  5. Adaptation for Micro-Architecture DSE
  6. Multi-Fidelity Reinforcement Learning
  7. Low-fidelity Training with Model-based RL
  8. High-fidelity Training
  9. Experiment
  10. Evaluation of Application-Specific Usage
  11. Evaluation of General-Purpose Usage
  12. Interpretability
  13. Conclusion

Figures (7)

  • Figure 1: The framework of our proposed methods.
  • Figure 2: Comparison between black-box methods and Fuzzy rule-based system.
  • Figure 3: Structure of Fuzzy Neural Networks
  • Figure 4: FNN with multi-fidelity RL.
  • Figure 5: Comparison with baselines.
  • ...and 2 more figures