Skip to main content

Ehsan Shareghi is an assistant professor in the Department of Data Science and Artificial Intelligence of Monash, and an affiliated lecturer of University of Cambridge. Prior to this, he was an associate lecturer at the Department of Electrical and Electronic Engineering of UCL, and before that a postdoc in the Language Technology Lab at University of Cambridge for a few years. I lead a small team of grad and undergrad students working on predictive models for language (text and speech). During my PhD and early parts of Postdoc, I worked on Bayesian nonparametric models and Bayesian learning of neural models and their application in language modelling and parsing. Since 2019 I have been working on various tasks and settings where training data is not sufficient for generalization at test or deployment due to several reasons; insufficient training data, knowledge-intensive nature of the task, absence of symbolic constraints. While in principle the solutions we investigate are agnostic to the underlying models, we try to leverage pretrained large models as much as possible. Hence, in recent years we have been also investigating various shortcomings of such models.

Research Areas in Language and Text.

  • Deep generative models (e.g., generation, semi-supervised learning, sparsity and disentanglement)
  • Self-supervised learning (e.g., applications in text, speech, and graph)
  • Utilising graphs for knowledge intensive tasks (e.g., graph-to-text generation, applications in biomedical domain)
  • Probing and augmenting pretrained large models (e.g., representational and computational efficiency, reliability)
  • Bayesian learning and inference (e.g., applications in training with small data)