Machine Learning-Based Pruning Technique for Low Power Approximate Computing

Approximate Computing is a low power achieving technique that offers an additional degree of freedom to design digital circuits. Pruning one the types approximate circuit which removes logic gates or wires in reduce consumption with minimal insertion error. In this work, novel machine learning (ML) -based pruning introduced The machine-learning algorithm random forest decision tree used prune nodes selectively based on their input pattern. addition, error compensation value added original output rate. Experimental results proved efficiency proposed terms area, and Compared conventional pruning, ML achieves 32% 26% area delay reductions 8*8 multiplier implementation. Low image processing algorithms are essential various applications like compression enhancement algorithms. For real-time evaluation, optimized applied discrete cosine transform (DCT). It basic element video applications. benchmark images show very good peak signal-to-noise ratio (PSNR) considerable amount energy savings compared other methods.