List your company
ResearchFormerly WizWand

Language Modeling on Text8 word-level (test)

99.9PerplexityNeural cache model
96.416119.933143.45166.967Mar 31, 2015Jul 12, 2015Oct 24, 2015Feb 5, 2016May 19, 2016Aug 31, 2016Dec 13, 2016
Updated 8mo ago

Evaluation results

MethodLinks
Neural cache modelbackbone=LSTM-10242016.1299.9
LSTM-1024hidden units=1024, backbone=LSTM, implementation=our implementation2016.12121.8
MemN2N# of hidden=500, # of hops=7, memory size=1002015.03147
MemNNbackbone=MemNN2016.12147
LSTM# of hidden=5002015.03154
MemN2N# of hidden=500, # of hops=5, memory size=1002015.03154
MemN2N# of hidden=500, # of hops=6, memory size=1502015.03154
MemN2N# of hidden=500, # of hops=6, memory size=1002015.03155
LSTM-500hidden units=500, backbone=LSTM2016.12156
MemN2N# of hidden=500, # of hops=6, memory size=1252015.03157
MemN2N# of hidden=500, # of hops=6, memory size=752015.03158
SCRN# of hidden=5002015.03161
SCRNNbackbone=SCRNN2016.12161
MemN2N# of hidden=500, # of hops=4, memory size=1002015.03162
MemN2N# of hidden=500, # of hops=6, memory size=252015.03163
MemN2N# of hidden=500, # of hops=6, memory size=502015.03166
MemN2N# of hidden=500, # of hops=3, memory size=1002015.03178
RNN# of hidden=5002015.03184
MemN2N# of hidden=500, # of hops=2, memory size=1002015.03187