Silhouette: Toward Performance-Conscious and Transferable CPU Embeddings

TL;DR

Silhouette introduces CPU embeddings learned from performance data to enhance transfer learning across diverse datasets, improving accuracy and transferability.

Contribution

The paper presents Silhouette, a novel method for learning CPU embeddings from performance data to facilitate transfer learning across different datasets.

Findings

Embeddings improve transfer learning accuracy.

Method enables transfer between datasets of varying types and sizes.

Performance data-driven embeddings enhance model generalization.

Abstract

Learned embeddings are widely used to obtain concise data representation and enable transfer learning between different data sets and tasks. In this paper, we present Silhouette, our approach that leverages publicly-available performance data sets to learn CPU embeddings. We show how these embeddings enable transfer learning between data sets of different types and sizes. Each of these scenarios leads to an improvement in accuracy for the target data set.