Cov Lus Cog Tseg Kev Ruaj Ntseg ua haujlwm los ntawm kev sib tw cov ntaub ntawv ntawm cov txiaj ntsig tshwj xeeb (xws li cov nqi me me zero). … L1 normalization ntxiv ib qho kev nplua L1 sib npaug rau tus nqi tag nrho ntawm qhov loj ntawm coefficients. Hauv lwm lo lus, nws txwv qhov loj ntawm cov coefficients.
L1 thiab L2 ua haujlwm li cas?
Qhov sib txawv tseem ceeb ntawm L1 thiab L2 tsis tu ncua yog qhov L1 tsis tu ncua sim kwv yees qhov nruab nrab ntawm cov ntaub ntawvthaum L2 tsis tu ncua sim kwv yees qhov nruab nrab ntawm cov ntaub ntawv rau tsis txhob overfitting. … Tus nqi ntawd kuj yuav yog qhov nruab nrab ntawm cov ntaub ntawv faib ua lej.
YPuas L1 lossis L2 normalization zoo dua?
Los ntawm qhov ua tau zoo, L1 zoo li txo qis coefficients mus rau xoom whereas L2 nyhav kom qis coefficients sib npaug. L1 yog li ntawd muaj txiaj ntsig zoo rau kev xaiv tshwj xeeb, vim peb tuaj yeem tso cov kev hloov pauv uas cuam tshuam nrog cov coefficients mus rau xoom. L2, ntawm qhov tod tes, muaj txiaj ntsig zoo thaum koj muaj cov yam ntxwv sib xyaw ua ke / codependent.
Regularizer ua haujlwm li cas?
Regularization ua haujlwm los ntawm ntxiv rau lub txim lossis lub sijhawm nyuaj lossis lub sijhawm luv luv nrog cov seem ntawm cov Squares (RSS) rau cov qauv nyuaj . β0, β1, … β sawv cev rau qhov kev kwv yees kwv yees sib txawv rau qhov sib txawv lossis tus kwv yees (X), uas piav qhia qhov hnyav lossis qhov loj ntawm qhov txuas nrog rau cov yam ntxwv, raws li.
L1 kev ua kom tsis tu ncua li cas txo Kev Tshaj Tawm?
L1 normalization, tseem hu ua L1 tus qauv lossis Lasso (hauv cov teeb meem rov qab), sib ntaus overfitting los ntawm kev txo qis ntawm qhov tsis sib xws ntawm 0.