Machine Learning: Finding pre-trained models for transfer learning

Abstract:

Training models from scratch requires way too much data, time, and computing power (or money) to be a practical option. This is why transfer learning has become such a common practice: by starting with models trained on related problems, you are saving time and achieving good results with little data. Now, where do you find such models? In this workshop, we will have a look at some of the most popular pre-trained models repositories and libraries (Model Zoo, PyTorch Hub, and Hugging Face); see how you can also search models in the literature and on GitHub, and finally learn how to import models into PyTorch.

Prerequisites:

If you want to follow the hands-on part of this workshop, please make sure to have an up-to-date version of PyTorch (https://pytorch.org/get-started/locally/) on your laptop.

Location:

Walter C. Koerner Library, UBC Library Research Commons Presentation Room (548/552)

Presenter:

Marie-Hélène Burle. An evolutionary and behavioural ecologist by training, Software/Data Carpentry instructor, and open source advocate, Marie-Hélène Burle develops and delivers training for researchers on high-performance computing tools (R, Python, Julia, Git, Bash scripting, machine learning, parallel scientific programming, and HPC) for Simon Fraser University and the Digital Research Alliance of Canada.