Timezone: »

 
Poster
Efficient Rematerialization for Deep Networks
Ravi Kumar · Manish Purohit · Zoya Svitkina · Erik Vee · Joshua Wang

Thu Dec 12 05:00 PM -- 07:00 PM (PST) @ East Exhibition Hall B + C #197
in Deep Learning -- Optimization for Deep Networks »

When training complex neural networks, memory usage can be an important bottleneck. The question of when to rematerialize, i.e., to recompute intermediate values rather than retaining them in memory, becomes critical to achieving the best time and space efficiency. In this work we consider the rematerialization problem and devise efficient algorithms that use structural characterizations of computation graphs---treewidth and pathwidth---to obtain provably efficient rematerialization schedules. Our experiments demonstrate the performance of these algorithms on many common deep learning models.

Keywords: Optimization for Deep Networks · Computational Complexity · Deep Learning · Algorithms; Optimization -> Combinatorial Optimization; Theory
[ Paper]  [ Poster] 

Author Information

Ravi Kumar (Google)
Manish Purohit (Google)
Zoya Svitkina (Google)
Erik Vee (Google)
Joshua Wang (Google Research)

More from the Same Authors