Deep Model Compression and Inference Speedup of Sum-Product Networks on Tensor Trains for Matlab©/Octave©
This package contains Matlab/Octave code for converting a trained SPN to conpact tSPN.
- Functions
- demo
Demonstrates the usage of the tspn_iden algorithm in converting trained SPNs to tSPNs.
- [core,nz,data,testdata]=tspn_iden(tensor,weight,sample_train,sample_test,opts)
Converts an SPN to its tSPN.
- nonrepeated.m
Finds non-repeated samples.
- findnonsample.m
Finds non-samples (negative sampling).
- inference = cpSPNinf(tensor,weight,samples)
Computes the inference for canonical polyadic terms.
- Reference
"Deep Model Compression and Inference Speedup of Sum-Product Networks on Tensor Trains"
Authors: Ching-Yun Ko, Cong Chen, Yuke Zhang, Kim Batselier, Ngai Wong