Unsupervised and semi-supervised embeddings for word sequences

EDIC candidacy exam
Exam president: Prof. Volkan Cevher
Thesis advisor: Prof. Martin Jaggi
Co-examiner: Prof. Boi Faltings

Abstract
The recent tremendous success of unsupervised word embeddings in a multitude of applications raises the obvious question if similar methods could be derived to improve embeddings (i.e. semantic representations) of word sequences as well. We plan to use a variety of machine learning methods as well as try to devise new model formulations to generate robust representations of word sequences in an unsupervised/semi-supervised fashion. We also plan to explore the mathematical underpinnings behind these models.

Background papers
Skip-Thought Vectors, Kiros et al. Arxiv.org.
Distributed Representations of Sentences and Documents, Le and Mikolov. Stanford edu.
Jointly optimizing word representations for lexical and sentential tasks with the C-PHRASE model, Pham et al.