Register now After registration you will be able to apply for this opportunity online.
Master thesis proposal: Uncertainty Aware Large Language Models
Generative Pre-trained Transformers have enabled an unprecedented level of performance in
Natural Language Processing tasks via Large Language Models (LLMs). Yet, a major aspect
that is hindering their use in critical applications is their lack of reliability. Hence, one approach
to address this question is through the lens of uncertainty quantification.
Nonetheless, current work either requires the retraining of the models, which is not
suitable for LLMs, or incur a high computational cost. To cope with that, the goal
of this project is to leverage recently proposed probes for LLMs for tractable Bayesian
learning and uncertainty quantification for various downstream tasks.
Keywords: Large Language Models (LLMs), Computer Science, Machine Learning, Probabilistic
Artificial Intelligence
Prerequisites: Deep learning, Bayesian statistics, Strong Python programming skills, knowledge
of Pytorch library, Strong interest in LLMs
Prerequisites: Deep learning, Bayesian statistics, Strong Python programming skills, knowledge of Pytorch library, Strong interest in LLMs
The goal of this project is to leverage recently proposed probes for LLMs for tractable Bayesian
learning and uncertainty quantification for various downstream tasks.
The goal of this project is to leverage recently proposed probes for LLMs for tractable Bayesian learning and uncertainty quantification for various downstream tasks.
Dr. Hossein Gorji: mohammadhossein.gorji@empa.ch
Ramzi Dakhmouche (Phd Student EPFL): ramzi.dakhmouche@empa.ch
Dr. Hossein Gorji: mohammadhossein.gorji@empa.ch Ramzi Dakhmouche (Phd Student EPFL): ramzi.dakhmouche@empa.ch
Bibliography related to Master thesis proposal.pdf
