A Transistor Operations Model for Deep Learning Energy Consumption Scaling Law

Deep Neural Networks (DNN) has transformed the automation of a wide range industries and finds increasing ubiquity in society. The high complexity DNN models its widespread adoption led to global energy consumption doubling every 3-4 months. Current measures largely monitor system or make linear assumptions models. former approach captures other unrelated anomalies, whilst latter does not accurately reflect nonlinear computations. In this paper, we are first develop bottom-up Transistor Operations (TOs) expose role non-linear activation functions neural network structure. As there will be inevitable measurement errors at core level, statistically model scaling laws as opposed absolute values. We offer for both feedforward DNNs convolution networks (CNNs) on variety data sets hardware configurations - achieving 93.6% 99.5% precision. This outperforms existing FLOPs-based methods our TOs method can further extended