<div dir="ltr"><div class="gmail_default"><div class="gmail_default"><font color="#000000"><span style="letter-spacing:0.2px;white-space:pre-wrap"><font face="georgia, serif" style=""><b style="">When:    </b>Monday, <b style="">January 9th at </b></font></span><b style="font-family:georgia,serif;letter-spacing:0.2px;white-space:pre-wrap"> <span style="background-color:rgb(255,255,0)">11:30AM CT</span></b></font></div><div class="gmail_default"><b style="letter-spacing:0.2px;white-space:pre-wrap"><font face="georgia, serif" color="#000000"><br></font></b></div><div class="gmail_default"><font face="georgia, serif" color="#000000"><b style="letter-spacing:0.2px;white-space:pre-wrap">Where:    </b>Talk will be given<span style="background-color:rgb(255,255,0)"> </span><span style="background-color:rgb(255,255,0)"><font style="font-weight:bold"><u>live, in-person</u></font><font style="font-weight:bold"> </font></span>at</font></div><p class="MsoNormal" style="margin:0in;line-height:normal;background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial"><font color="#000000" face="georgia, serif">                    TTIC, 6045 S. Kenwood Avenue</font></p><p class="MsoNormal" style="margin:0in;line-height:normal;background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial"><font color="#000000" face="georgia, serif">                    5th Floor, Room 530<b>  </b></font></p><p class="MsoNormal" style="margin:0in;line-height:normal;background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial"><b style="letter-spacing:0.2px;white-space:pre-wrap"><font face="georgia, serif" color="#000000"><br></font></b></p><p class="MsoNormal" style="margin:0in;line-height:normal;background-image:initial;background-position:initial;background-size:initial;background-repeat:initial;background-origin:initial;background-clip:initial"><font face="georgia, serif" color="#000000"><b style="letter-spacing:0.2px;white-space:pre-wrap">Virtually: </b><span style="letter-spacing:0.2px;white-space:pre-wrap">via Panopto (</span><a href="https://uchicago.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=a8661e99-062f-42ff-9fd5-af780145c4d5" style="letter-spacing:0.2px;white-space:pre-wrap" target="_blank">Livestream</a><span style="letter-spacing:0.2px;white-space:pre-wrap">)</span><br></font></p><div class="gmail_default"><b style="letter-spacing:0.2px;white-space:pre-wrap"><font face="georgia, serif" color="#000000"><br></font></b></div><div class="gmail_default"><span style="letter-spacing:0.2px;white-space:pre-wrap"><font face="georgia, serif" color="#000000"><b>Who:     </b>Leon Bottou, Facebook</font></span></div><div class="gmail_default"><b style="letter-spacing:0.2px;white-space:pre-wrap"><font face="georgia, serif" color="#000000"><br></font></b></div><div class="gmail_default"><font color="#000000"><font face="georgia, serif"><b style="letter-spacing:0.2px;white-space:pre-wrap">Title:     </b>Out-of-distribution generalization, causation, and features<br style="letter-spacing:0.2px;white-space:pre-wrap"><br style="letter-spacing:0.2px;white-space:pre-wrap"><b style="letter-spacing:0.2px;white-space:pre-wrap">Abstract:</b><span style="letter-spacing:0.2px;white-space:pre-wrap"> </span>Out-of-distribution generalization is only possible when we assume that </font><span style="font-family:georgia,serif">there is a connection between the training and testing distribution. For </span><span style="font-family:georgia,serif">instance, we can make the assumption that these distributions share a </span><span style="font-family:georgia,serif">same causal model but differ because of changes in the distribution of </span><span style="font-family:georgia,serif">uncontrolled variables. But how then to recover enough information about </span><span style="font-family:georgia,serif">this causal model and these uncontrolled variables.  Learning causal </span><span style="font-family:georgia,serif">invariances (Arjovsky et al. 2019) offers an interesting workaround but </span><span style="font-family:georgia,serif">is plagued by optimization problems that severely limit its impact.  We </span><span style="font-family:georgia,serif">propose to first initialize the networks with a rich representation containing</span><span style="font-family:georgia,serif"> a palette of potentially useful features, ready to be used by </span><span style="font-family:georgia,serif">even simple models. Such a representation is constructed with a </span><span style="font-family:georgia,serif">succession of specially crafted training episodes. This approach consistently helps six OoD methods achieve top performance </span>on ColoredMNIST<span style="font-family:georgia,serif"> benchmark. The same technique substantially outperforms comparable results on the Wilds Camelyon17 task, eliminates the high result variance that plagues other methods, and makes hyperparameter tuning and model selection more reliable.</span></font></div><div><font face="georgia, serif" color="#000000"><br></font></div><div><div class="gmail_default"><font color="#000000"><font face="georgia, serif"><b>Bio: </b>Léon Bottou received the Diplôme d'Ingénieur de l'École Polytechnique </font><span style="font-family:georgia,serif">(X84) in 1987, the Magistère de Mathématiques Fondamentales et </span><span style="font-family:georgia,serif">Appliquées et d'Informatique from École Normale Supérieure in 1988, and </span><span style="font-family:georgia,serif">a Ph.D. in Computer Science from Université de Paris-Sud in 1991. His </span><span style="font-family:georgia,serif">research career took him to AT&T Bell Laboratories, AT&T Labs Research, </span><span style="font-family:georgia,serif">NEC Labs America, and Microsoft Research. He joined Facebook AI Research </span><span style="font-family:georgia,serif">in 2015. The long-term goal of Léon Bottou's research is to understand </span><span style="font-family:georgia,serif">and replicate human-level intelligence. Because this goal requires </span><span style="font-family:georgia,serif">conceptual advances that cannot be anticipated, Leon's research has </span><span style="font-family:georgia,serif">followed many practical and theoretical turns: neural networks </span><span style="font-family:georgia,serif">applications in the late 1980s, stochastic gradient learning algorithms </span><span style="font-family:georgia,serif">and statistical properties of learning systems in the early 1990s, </span><span style="font-family:georgia,serif">computer vision applications with structured outputs in the late 1990s, </span><span style="font-family:georgia,serif">theory of large scale learning in the 2000s. During the last few years, </span><span style="font-family:georgia,serif">Léon Bottou's research aims to clarify the relation between learning and </span><span style="font-family:georgia,serif">reasoning, with more and more focus on the many aspects of causation </span><span style="font-family:georgia,serif">(inference, invariance, reasoning, affordance, and intuition.</span></font></div><div class="gmail_default" style="font-size:small;font-family:Arial,Helvetica,sans-serif"><font face="georgia, serif"><b><br></b></font></div><div class="gmail_default" style="font-size:small;font-family:Arial,Helvetica,sans-serif"><font face="georgia, serif"><b><br></b></font></div></div></div><div><div class="gmail_default" style="font-family:georgia,serif;font-size:small"><b>Host: <a href="mailto:nati@ttic.edu" target="_blank">Nati Srebro</a></b></div><br></div><div><br></div><div><br></div>-- <br><div dir="ltr" data-smartmail="gmail_signature"><div dir="ltr"><b style="background-color:rgb(255,255,255)"><font color="#3d85c6">Brandie Jones </font></b><div><div><div><font color="#3d85c6"><b><i>Executive </i></b></font><b style="color:rgb(61,133,198)"><i>Administrative Assistant</i></b></div></div><div><span style="background-color:rgb(255,255,255)"><font color="#3d85c6">Toyota Technological Institute</font></span></div><div><span style="background-color:rgb(255,255,255)"><font color="#3d85c6">6045 S. Kenwood Avenue</font></span></div><div><span style="background-color:rgb(255,255,255)"><font color="#3d85c6">Chicago, IL  60637</font></span></div></div><div><span style="background-color:rgb(255,255,255)"><font color="#3d85c6"><a href="http://www.ttic.edu" target="_blank">www.ttic.edu</a> </font></span></div><div><font color="#3d85c6"><div style="background-color:rgb(238,238,238)">Working Remotely on Tuesdays</div></font></div></div></div></div>