<html><head><meta http-equiv="content-type" content="text/html; charset=us-ascii"></head><body style="overflow-wrap: break-word; -webkit-nbsp-mode: space; line-break: after-white-space;"><div><div><div>This is an announcement of Kevin Suk's MS Presentation</div><div><br></div><div>===============================================</div><div>Candidate: Kevin Suk</div><div><br></div><div>Date: Thursday, May 2, 2024</div><div><br></div><div>Time: 3 pm CT</div><div><br></div><div>Location: JCL 298</div><div><br></div><div>Title: <span style="caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0);">Probabilistic Contrastive Language-Image Pretraining</span></div><div><br></div><div>Abstract: <span style="caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0);">CLIP utilizes cosine similarities to perform contrastive learning. Although it is efficient, the cosine only captures the angle between two vectors, enforcing all vectors onto an unit hypersphere. This could limit the expressive power of the model as it might not optimally capture the relationships between different modalities in the vector space. Moreover, the cosine similarity is symmetric, failing to discriminate between classification of images with their corresponding captions and classification of captions with images. Hence, we aim to improve performance by directly utilizing the conditional probabilities $\mathbb{P}(\text{Image}|\text{Text})$, $\mathbb{P}(\text{Text}|\text{Image})$ to form a new asymmetric contrastive loss. Testing on multiple new and old benchmarks, we show that under resource constrained training conditions, Probabilistic CLIP is able to provide better or at-par performance compared to CLIP with a higher confidence level.</span></div><div><br></div><div>Advisors: <span style="caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0);">Greg Shakhnarovich</span></div><div><br></div><div>Committee Members: <span style="caret-color: rgb(0, 0, 0); color: rgb(0, 0, 0);">Greg Shakhnarovich, Michael Maire, Yuxin Chen</span></div></div></div><div><br></div><div><br></div><div>
<meta charset="UTF-8"><div dir="auto" style="caret-color: rgb(0, 0, 0); letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; overflow-wrap: break-word; -webkit-nbsp-mode: space; line-break: after-white-space;"><font face="AvenirNext-Regular" color="#8b0021"><span style="font-style: normal; font-size: 14px;"><b>Jessica Garza</b></span></font><br><font face="Avenir-Book" style="color: rgb(0, 0, 0);"><span style="font-size: 14px;">Assistant Director of Undergraduate Studies</span></font><br><font face="Avenir-Book" style="color: rgb(0, 0, 0);"><span style="font-size: 14px;">Department of Computer Science</span></font><br><font face="Avenir-Book" style="color: rgb(0, 0, 0);"><span style="font-size: 14px;">The University of Chicago</span></font><br><font face="Avenir-Book" style="color: rgb(0, 0, 0);"><span style="font-size: 14px;">John Crerar Library 374</span></font><br><font face="Avenir-Book" style="color: rgb(0, 0, 0);"><span style="font-size: 14px;">Office: (773) 702-2336</span></font></div>
</div>
<br></body></html>