Learning Classifier Tables for Autonomic Systems on Chip

Abstract

This paper introduces a new hardware-based machine learning building block – called Learning Classifier Table (LCT) – for the run-time reliability, performance and power optimization of future generations of Systems-on-Chip. LCT inherits concepts from the reinforcement learning techniques found in Learning Classifier Systems. Prediction weighted LCT rule evaluation is implemented on a clock cycle scale with low hardware complexity.