Bigram Language Modeling on Synthetic Random 50% initialization (val) benchmark leaderboard · SOTA2 Research

Bigram Language Modeling on Synthetic Random 50% initialization (val)

0.0005Avg JS DivergenceMixture KLs

Evaluation Results

Method	Links
Mixture KLs2023.05		0.0005	0.0001
MIXCE2023.05		0.0007	0.0001
Forward KL2023.05		0.0007	0.0001
Oracle MIXCE2023.05		0.0008	0.0001
Mixture KLs2023.05		0.0045	0.0003
Oracle MIXCE2023.05		0.0045	0.0004
Forward KL2023.05		0.0065	0.0006
MIXCE2023.05		0.0065	0.0006
Mixture KLs2023.05		0.0274	0.0002
Oracle MIXCE2023.05		0.0285	0.0009
Forward KL2023.05		0.0356	0.0014
MIXCE2023.05		0.0356	0.0014
Reverse KL2023.05		0.136	0
Jensen-Shannon2023.05		0.214	0
Mixture KLs2023.05		0.232	0.0009
Oracle MIXCE2023.05		0.234	0.0014
MIXCE2023.05		0.235	0.0015
Forward KL2023.05		0.239	0.0015
Mixture KLs2023.05		0.291	0.0009
Oracle MIXCE2023.05		0.292	0.0009
MIXCE2023.05		0.292	0.0009
Forward KL2023.05		0.293	0.0009
Reverse KL2023.05		0.429	0.0015
Jensen-Shannon2023.05		0.459	0.0006
Jensen-Shannon2023.05		0.474	0.0014
Jensen-Shannon2023.05		0.534	0.0007
Jensen-Shannon2023.05		0.553	0.001
Reverse KL2023.05		0.557	0.0004
Reverse KL2023.05		0.678	0
Reverse KL2023.05		0.685	0