<div dir="ltr"><div><div>Caros,<div><br></div><div><div>Na sexta-feira, 8 de Agosto às 14:00hs, teremos o seguinte seminário do Programa Interinstitucional de Pós-Graduação em Estatística (PIPGEs) da UFSCar-USP.<br></div><div><br></div><div><b>Nome</b>: Alek Fröhlich (Italian Institute of Technology)</div></div><div><p><strong>Título: </strong>Spectral Representation Learning for High-Dimensional Inference</p></div></div><div><div><b>Resumo: </b>In this talk, we explore how the truncated singular value decomposition
of conditional expectation operators yields useful representations for
problems such as regression, uncertainty quantification, and conditional
probability estimation. We introduce a learning algorithm reminiscent
of contrastive methods in computer vision and density ratio estimation,
and show how the quality of learned representations translates to
downstream inference tasks with statistical guarantees. We conclude with
a brief look at ongoing work on spectral representation learning in the
contexts of distributional symmetries, reinforcement learning, and
conditional independence testing.</div></div><div><div><p style="line-height:14.04px;margin-bottom:0.28cm;direction:ltr;background:transparent"><b>Local: </b>Sala 5 do CINA-UFSCar (São Carlos) </p></div><div>Todos são bem vindos!</div></div><br clear="all"></div><div><div dir="ltr" class="gmail_signature" data-smartmail="gmail_signature"><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div><div dir="ltr"><div> Rafael<br><br>--<br></div>Rafael Izbicki<br>Associate Professor | Vice Director of Graduate Studies<br>Department of Statistics<br>Federal University of São Carlos (UFSCar)<br><a href="https://rafaelizbicki.com/" target="_blank">https://rafaelizbicki.com/</a><br></div><div dir="ltr"><a href="https://small-research.github.io/website" target="_blank">https://small-research.github.io/website</a></div><div dir="ltr"><br><br></div></div></div></div></div></div></div></div></div></div></div></div></div></div>