<div dir="ltr">appears quite relevant to a few of us<br><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">---------- Forwarded message ---------<br>From: <strong class="gmail_sendername" dir="auto">Diane Stidle</strong> <span dir="auto"><<a href="mailto:stidle@andrew.cmu.edu">stidle@andrew.cmu.edu</a>></span><br>Date: Fri, Sep 25, 2020 at 12:40 PM<br>Subject: Thesis Proposal - Oct. 2, 2020 - Otilia Stretcu - Curriculum Learning<br>To: <a href="mailto:ml-seminar@cs.cmu.edu">ml-seminar@cs.cmu.edu</a> <<a href="mailto:ML-SEMINAR@cs.cmu.edu">ML-SEMINAR@cs.cmu.edu</a>>, Rich Caruana <<a href="mailto:rcaruana@microsoft.com">rcaruana@microsoft.com</a>><br></div><br><br>
<div>
<p><i><b>Thesis Proposal</b></i></p>
<p>Date: October 2, 2020<br>
Time: 3:30pm (EDT)<br>
Speaker: Otilia Stretcu</p>
<p>Zoom Meeting: <a href="https://cmu.zoom.us/j/94122713476?pwd=VlFqOWYrRkFQMnZaSE0vTXFtT3pRdz09" target="_blank">https://cmu.zoom.us/j/94122713476?pwd=VlFqOWYrRkFQMnZaSE0vTXFtT3pRdz09<br>
</a>Meeting ID: 941 2271 3476<br>
Passcode: 866793</p>
<p><b>Title: Curriculum Learning</b></p>
<p>Abstract:<br>
AI researchers often disagree about the best strategy to train a
machine learning system, but there is one belief that is generally
agreed upon: humans are still much better learners than machines.
Unlike AI systems, humans do not learn difficult new tasks (e.g.,
solving differential equations) from scratch, by looking at
independent and identically distributed examples of the task being
performed by someone else. Instead, new skills are often built
progressively, starting with easier tasks and gradually becoming
able to perform harder ones. Curriculum Learning (CL) is a line of
work that tries to incorporate this human approach to learning
into machine learning. In this thesis we aim to discover the
problem settings in which different forms of CL are beneficial,
and the types of benefits they provide. Our completed work in
machine translation and image classification already showcases two
different settings in which CL is successful. Next, we plan to
take this work further and tackle some problems that are even more
challenging for modern machine learning systems, such as function
composition and learning to do math with neural networks. If
successful, this work could help CL eventually become the standard
method for training systems, bringing machine learning one step
closer to human intelligence.</p>
<div dir="ltr"><br>
<b>
Thesis Committee</b>: </div>
<div dir="ltr">Tom Mitchell, Co-Chair</div>
<div dir="ltr">
Barnabás Póczos, Co-Chair</div>
<div dir="ltr">
Ruslan Salakhutdinov</div>
Rich Caruana, Microsoft Research
<p><br>
<br>
<br>
</p>
</div>
</div></div>