オートエンコーダ (Japanese Wikipedia)

Analysis of information sources in references of the Wikipedia article "オートエンコーダ" in Japanese language version.

refsWebsite

Global rank Japanese rank

5jmlr.org

low place

1toronto.edu

low place

jmlr.org

"autoencoder where Y is of the same dimensionality as X (or larger) can achieve perfect reconstruction simply by learning an identity mapping." Vincent. (2010). Stacked Denoising Autoencoders: Learning Useful Representations in a Deep Network with a Local Denoising Criterion.
"The criterion that representation Y should retain information about input X is not by itself sufficient to yield a useful representation." Vincent. (2010). Stacked Denoising Autoencoders: Learning Useful Representations in a Deep Network with a Local Denoising Criterion.
"a deterministic mapping from X to Y, that is, ... equivalently $q(Y|X;\theta )=\delta (Y-f_{\theta }(X))$ ... The deterministic mapping $f_{\theta }$ that transforms an input vector ${\boldsymbol {x}}$ into hidden representation ${\boldsymbol {y}}$ is called the encoder." Vincent. (2010). Stacked Denoising Autoencoders: Learning Useful Representations in a Deep Network with a Local Denoising Criterion.
" ${\boldsymbol {z}}=g_{\theta ^{'}}({\boldsymbol {y}})$ . This mapping $g_{\theta ^{'}}$ is called the decoder. ... In general ${\boldsymbol {z}}$ is not to be interpreted as an exact reconstruction of ${\boldsymbol {x}}$ , but rather in probabilistic terms as the parameters (typically the mean) of a distribution $p(X|Z={\boldsymbol {z}})$ " Vincent. (2010). Stacked Denoising Autoencoders: Learning Useful Representations in a Deep Network with a Local Denoising Criterion.
" $g_{\theta ^{'}}$ is called the decoder ... $Z=g_{\theta ^{'}}({\boldsymbol {y}})$ ... associated loss function $L({\boldsymbol {x}},{\boldsymbol {z}})$ ... $X|{\boldsymbol {z}}\sim N({\boldsymbol {z}},{\boldsymbol {\sigma }}^{2}{\boldsymbol {I}})$ ... This yields $L({\boldsymbol {x}},{\boldsymbol {z}})=L_{2}({\boldsymbol {x}},{\boldsymbol {z}})=C(\sigma ^{2})\|{\boldsymbol {x}}-{\boldsymbol {z}}\|^{2}$ ... This is the squared error objective found in most traditional autoencoders." Vincent. (2010). Stacked Denoising Autoencoders: Learning Useful Representations in a Deep Network with a Local Denoising Criterion.

toronto.edu

cs.toronto.edu

Geoffrey E. Hinton; R. R. Salakhutdinov (2006-07-28). “Reducing the Dimensionality of Data with Neural Networks”. Science 313 (5786): 504-507.