Machine Learning-Based Error Detection and Function Approximation for Reliable Computing

Ensuring reliability in modern computing systems requires efficient and accurate error detection mechanisms. This paper explores a machine learning based approach to error detection through accurate approximation of commonly used exponential and logarithmic functions. A hybrid error detection architecture is proposed, which combines a traditional computation module and a ML-based approximator running in parallel. Several regression models are evaluated based on their approximation accuracy and complexity. Experimental results show that polynomial regression offers the best compromise between precision and complexity, while MLP and SVR provide high accuracy with increased resource requirements.