Enhancing Edge Intelligence with Layer-wise Adaptive Precision and Randomized PCA

Edge intelligence is the ability of edge devices to carry out intelligent operations, such as object identification, speech recognition, or natural language processing, utilizing machine learning algorithms. The primary goal is to fix edge computing's problems and improve its performance. The main goal of this work is to apply RPCA to increase energy efficiency and reduce memory usage. The algorithm computes the covariance matrix of the centered data, finds the eigenvectors and eigenvalues of the covariance matrix, sorts the eigenvectors and eigenvalues in descending order of the eigenvalues, chooses the first set of eigenvectors, and projects the data onto the chosen eigenvectors. This article employs a technique known as layer-wise adaptive precision (LAP), which decreases the precision of activations in neural network layers that contribute less to output accuracy.