The result for coherent matrices are not surprising IMHO, because best performing methods("Bayesian" T_MSBL etc) are using additional information - which vectors are coherent and factor that into measurement model. The only surprising thing is that Smooth L0 sometimes works. IIRC it's based on something like Cauchy estimator, so it can explain why it more robust than L1. PS it it possible to turn off captcha? I have difficulty processing it - I'm not a bot )))
Liked this entry ? subscribe to the Nuit Blanche feed, there's more where that came from (You can also subscribe to Nuit Blanche by Email)
The result for coherent matrices are not surprising IMHO, because best performing methods("Bayesian" T_MSBL etc) are using additional information - which vectors are coherent and factor that into measurement model. The only surprising thing is that Smooth L0 sometimes works. IIRC it's based on something like Cauchy estimator, so it can explain why it more robust than L1.
ReplyDeletePS it it possible to turn off captcha? I have difficulty processing it - I'm not a bot )))
Serguey,
ReplyDeleteI am sorry I am going to put back the capcha as I am getting flooded with spam. Sorry.
Igor.
From a quick glance to the experiment code I can say that comparison is not exactly fair, so the results are not surprising.
ReplyDeleteUlugbek,
ReplyDeleteCan you expand on that ?
Igor.