?url_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Adc&rft.title=CausalVAE%3A+Disentangled+Representation+Learning+via+Neural+Structural+Causal+Models&rft.creator=Yang%2C+Mengyue&rft.creator=Liu%2C+Furui&rft.creator=Chen%2C+Zhitang&rft.creator=Shen%2C+Xinwei&rft.creator=Hao%2C+Jianye&rft.creator=Wang%2C+Jun&rft.description=Learning+disentanglement+aims+at+finding+a+low+dimensional+representation+which+consists+of+multiple+explanatory+and+generative+factors+of+the+observational+data.+The+framework+of+variational+autoencoder+(VAE)+is+commonly+used+to+disentangle+independent+factors+from+observations.+However%2C+in+real+scenarios%2C+factors+with+semantics+are+not+necessarily+independent.+Instead%2C+there+might+be+an+underlying+causal+structure+which+renders+these+factors+dependent.+We+thus+propose+a+new+VAE+based+framework+named+CausalVAE%2C+which+includes+a+Causal+Layer+to+transform+independent+exogenous+factors+into+causal+endogenous+ones+that+correspond+to+causally+related+concepts+in+data.+We+further+analyze+the+model+identifiabitily%2C+showing+that+the+proposed+model+learned+from+observations+recovers+the+true+one+up+to+a+certain+degree.+Experiments+are+conducted+on+various+datasets%2C+including+synthetic+and+real+word+benchmark+CelebA.+Results+show+that+the+causal+representations+learned+by+CausalVAE+are+semantically+interpretable%2C+and+their+causal+relationship+as+a+Directed+Acyclic+Graph+(DAG)+is+identified+with+good+accuracy.+Furthermore%2C+we+demonstrate+that+the+proposed+CausalVAE+model+is+able+to+generate+counterfactual+data+through+%E2%80%9Cdo-operation%E2%80%9D+to+the+causal+factors.&rft.subject=Science+%26+Technology%2C+Technology%2C+Computer+Science%2C+Artificial+Intelligence%2C+Imaging+Science+%26+Photographic+Technology%2C+Computer+Science&rft.publisher=IEEE&rft.date=2021-11-13&rft.type=Proceedings+paper&rft.language=eng&rft.source=+++++In%3A++2021+IEEE%2FCVF+Conference+on+Computer+Vision+and+Pattern+Recognition+(CVPR).++(pp.+pp.+9588-9597).++IEEE%3A+Nashville%2C+TN%2C+USA.+(2021)+++++&rft.format=application%2Fpdf&rft.identifier=https%3A%2F%2Fdiscovery.ucl.ac.uk%2Fid%2Feprint%2F10142668%2F1%2F2004.08697.pdf&rft.identifier=https%3A%2F%2Fdiscovery.ucl.ac.uk%2Fid%2Feprint%2F10142668%2F&rft.rights=open