A computational-graph partitioning method for training memory-constrained DNNs.
Fareed QararyahMohamed WahibDoga DikbayirMehmet Esat BelviranliDidem UnatPublished in: Parallel Comput. (2021)
Keyphrases
- computational power
- training examples
- training set
- graph representation
- graph model
- training process
- mathematical programming
- graph theory
- graph structure
- memory usage
- directed graph
- graph theoretic
- memory requirements
- computing power
- random access
- memory space
- memory size
- weighted graph
- spanning tree
- limited memory
- training algorithm
- link analysis
- clustering algorithm
- bipartite graph
- graph matching
- main memory
- training samples
- online learning
- probabilistic model
- feature space
- training data
- decision trees