Design Space Exploration of Approximate Computing Techniques with a Reinforcement Learning Approach

TL;DR

The paper tackles the challenge of selecting appropriate AxC techniques to balance accuracy, power, and execution time. It proposes an RL-based, multi-objective Design Space Exploration framework that models approximate operators and variable selections and uses Q-learning to optimize trade-offs. Experimental evaluation on Matrix Multiplication and FIR benchmarks with EvoApproxLib operators shows the method can produce configurations that satisfy target thresholds and reveals task-dependent learnability. The work demonstrates a promising direction for automated AxC design and outlines clear avenues for improving learning strategies and generalization.

Abstract

Approximate Computing (AxC) techniques have become increasingly popular in trading off accuracy for performance gains in various applications. Selecting the best AxC techniques for a given application is challenging. Among proposed approaches for exploring the design space, Machine Learning approaches such as Reinforcement Learning (RL) show promising results. In this paper, we proposed an RL-based multi-objective Design Space Exploration strategy to find the approximate versions of the application that balance accuracy degradation and power and computation time reduction. Our experimental results show a good trade-off between accuracy degradation and decreased power and computation time for some benchmarks.

Figures (4)

• Figure 1: RL environment: at each step selects an approximated version of the benchmark to execute and calculates the parameters to return the new observations and rewards to the RL agent.
• Figure 2: Exploration outcomes evolution through for Matrix Multiplication (10x10).
• Figure 3: Exploration outcomes evolution through for FIR (100 samples).
• Figure 4: Average reward evolution for the Matrix multiplication (10x10) and FIR (100 samples).