Computer Science Colloquium – April Shen ’13
Fri, September 13th, 2019
2:30 pm - 4:00 pm
- This event has passed.
“Thinking Outside the Box: Non-Geometric Perspectives on Semantic Similarity”
Representation learning has been an important component of the modern deep learning renaissance. These representations can encode rich notions of semantic similarity between objects, which makes them useful for many different applications without necessarily requiring labelled task-specific datasets. However, this research often implicitly assumes a geometric view of these representations, seeing them as vectors in space and comparing them using geometric notions like angle or distance. But sometimes, questioning our assumptions and taking a different perspective can be illuminating. In this talk I’ll introduce some non-geometric perspectives of representations, including ones using fuzzy sets and statistical correlation measures, and focusing especially on representations of words and sentences. These approaches can lead in directions that wouldn’t otherwise be obvious and result in simple yet effective methods of measuring semantic similarity.
April Shen is a Machine Learning Engineer at Babylon Health in London, UK, where she contributes to natural language processing and machine learning research relevant to healthcare and other medical applications. Her interests fall mostly in deep learning and representation learning, and in particular the issues that come up when applying these techniques in practice, including safety and trust. She graduated from Williams in 2013 and was advised by Andrea Danyluk.
Tue, January 21st, 2020
Wed, January 22nd, 2020