This work reveals that memory-intensive computation is a rising performance-critical factor in
recent machine learning models. Due to a unique set of new challenges, existing ML …

The rapidly-changing deep learning landscape presents a unique opportunity for building
inference accelerators optimized for specific datacenter-scale workloads. We propose Full …

Today's auto-tuners (eg, AutoTVM, Ansor) generate efficient tensor programs by navigating
a large search space to identify effective implementations, but they do so with opaque …

We show in this work that memory intensive computations can result in severe performance
problems due to off-chip memory access and CPU-GPU context switch overheads in a wide …

This article proposes DisCo, an automatic deep learning compilation module for data-
parallel distributed training. Unlike most deep learning compilers that focus on training or …

The strong demand for efficient and performant deployment of Deep Learning (DL)
applications prompts the rapid development of a rich DL ecosystem. To keep up with this fast …

Designing efficient deep learning architectures is a challenging task that requires balancing
performance and hardware efficiency. Neural Architecture Search (NAS) has emerged as a …

Deep learning has achieved remarkable success across various domains, leading to the
development of increasingly complex and resource-intensive models. For that, these models …

Today, machine learning offers a variety of services in the industry; including research,
translation, recommendation systems and security. Deep learning in particular has led to …

Abstract Machine learning compilers rely on making optimized decisions in order to
generate efficient code for a given computation graph. Many of these decision making …