M
Marisa Kirisame
Researcher at University of Washington
Publications - 7
Citations - 207
Marisa Kirisame is an academic researcher from University of Washington. The author has contributed to research in topics: Compiler & Relay. The author has an hindex of 5, co-authored 6 publications receiving 101 citations.
Papers
More filters
Proceedings ArticleDOI
Relay: a new IR for machine learning frameworks
Jared Roesch,Steven Lyubomirsky,Logan Weber,Josh Pollock,Marisa Kirisame,Tianqi Chen,Zachary Tatlock +6 more
TL;DR: Relay as mentioned in this paper is a purely functional, statically-typed language with the goal of balancing efficient compilation, expressiveness, and portability for machine learning models across an array of heterogeneous hardware devices.
Proceedings ArticleDOI
Relay: A New IR for Machine Learning Frameworks
Jared Roesch,Steven Lyubomirsky,Logan Weber,Josh Pollock,Marisa Kirisame,Tianqi Chen,Zachary Tatlock +6 more
TL;DR: This work proposes a new high-level intermediate representation (IR) called Relay, being designed as a purely-functional, statically-typed language with the goal of balancing efficient compilation, expressiveness, and portability.
Posted Content
Dynamic Tensor Rematerialization
Marisa Kirisame,Steven Lyubomirsky,Altan Haan,Jennifer Brennan,Mike He,Jared Roesch,Tianqi Chen,Zachary Tatlock +7 more
TL;DR: Dynamic Tensor Rematerialization is presented, a greedy online algorithm for heuristically checkpointing arbitrary models and it is proved it can train an $N$-layer feedforward network on an $\Omega(\sqrt{N})$ memory budget with only $\mathcal{O}(N)$ tensor operations.
Posted Content
Relay: A High-Level Compiler for Deep Learning
Jared Roesch,Steven Lyubomirsky,Marisa Kirisame,Logan Weber,Josh Pollock,Luis Vega,Ziheng Jiang,Tianqi Chen,Thierry Moreau,Zachary Tatlock +9 more
TL;DR: Relay's design demonstrates how a unified IR can provide expressivity, composability, and portability without compromising performance.
Posted Content
Relay: A High-Level IR for Deep Learning.
Jared Roesch,Steven Lyubomirsky,Marisa Kirisame,Josh Pollock,Logan Weber,Ziheng Jiang,Tianqi Chen,Thierry Moreau,Zachary Tatlock +8 more
TL;DR: The functional, statically-typed Relay IR unifies and generalizes existing DL IRs and can express state-of-the-art models and can eliminate abstraction overhead and target new hardware platforms.